Embedded systems

Abstract

The system may include a data classification module configured to categorize data into classified data based on predefined confidentiality levels and regulatory compliance requirements. The system may include an access control module configured to manage permissions for different user roles within the enterprise and to grant access to classified data according to confidentiality levels and regulatory compliance requirements. The system may include a data formatting module configured to transform classified data into formatted data in customized display formats for various corporate departments. The system may include an integration module configured to interface with at least one of an enterprise resource planning (ERP) system or a customer relationship management (CRM) system to retrieve and categorize data. The system may include a user interface module that presents formatted data within a host application and provides a seamless user experience in accessing an embedded marketplace.

Description

[Technical Field]

[0001] [Cross-reference with related applications] This application claims priority to U.S. Provisional Patent Application No. 63 / 450,638, filed on March 7, 2023. This application claims priority to U.S. Provisional Patent Application No. 63 / 535,741, filed on August 31, 2023. This application claims priority to U.S. Provisional Patent Application No. 63 / 610,890, filed on December 15, 2023. This application claims priority to U.S. Provisional Patent Application No. 63 / 621,548, filed on January 16, 2024. This application claims priority to U.S. Provisional Patent Application No. 63 / 625,605, filed on January 26, 2024. This application claims priority to U.S. Provisional Patent Application No. 63 / 461,802, filed on April 25, 2023. Each of the patent applications mentioned above is incorporated herein by reference as if it were fully described herein in its entirety.

[0002] This disclosure relates to embedded marketplaces, and more specifically, to transaction systems that include embedded marketplaces. [Background technology]

[0003] The exponential increase in connectivity and intelligence across all types of devices has led to an unprecedented expansion in the scale and granularity of data worldwide, giving rise to entirely new types of data that can be used to enable and enhance digital transactions in all kinds of marketplaces. This expansion presents new challenges in analyzing and interpreting fractally expanding data layers to derive insights, as well as regulatory and business requirements to understand and address the transactions, transactional entities, and all companies, individuals, and AI intermediaries operating or interacting on the data.

[0004] Some transactions are related to marketplaces. Traditional marketplaces often require interaction with a specific location. This location may be the geographical location of a physical marketplace, or it may be a designated IP address or application for a specific product or vendor.

[0005] Marketplaces offer a range of critical functions to stakeholders, including the ability to find counterparties to agree on transactions involving diverse asset classes. Exchange transactions, among other things, enable the release of liquidity, the execution of financial strategies (such as arbitrage), risk management (such as options and futures contracts), capital aggregation, value conversion between asset classes, participation in trading profits, behavioral impact, and the acquisition of insights (such as data streams related to transactions). Successful marketplaces like the New York Stock Exchange (NYSE) and the Chicago Mercantile Exchange (CME) are fundamental components of the global economy, with new exchanges regularly emerging for new categories. Exchanges increasingly rely on the capabilities of information technology infrastructure for a wide range of core functions related to trading, offering, execution, reporting, analysis, matching, and other functions, including distributed storage, caching, high-speed networks, algorithmic trading, big data, data integration, modeling and analytics, robotic process automation, distributed ledger technology (DLT), smart contracts, real-time data collection, search, and asset digitization. In this technology field, there is a need to provide intelligent market orchestration that addresses a broad and expanding range of asset classes and involves an increasingly diverse group of stakeholders. [Overview of the Initiative]

[0006] In embodiments, the technology described herein relates to a computer-implemented system that provides an embedded marketplace within a host application, the system comprising: a data classification module configured to classify data into classified data based on predefined confidentiality levels and regulatory compliance requirements; an access control module configured to manage permissions for different user roles within an enterprise and to grant access to the classified data in accordance with confidentiality levels and regulatory compliance requirements; a data formatting module configured to convert the classified data into formatted data in a display format customized for various corporate departments; an integration module configured to interface with at least one of an enterprise resource planning (ERP) system and a customer relationship management (CRM) system to retrieve and classify data; and a user interface module configured to display the formatted data within a host application and to provide a seamless user experience in accessing the embedded marketplace.

[0007] In one embodiment, the host application for embedding the marketplace is an Enterprise Resource Planning (ERP) system, and the data classification module is further configured to classify financial data, supply chain data, and human resource data and selectively present them to authorized users. In another embodiment, the host application for embedding the marketplace is a Customer Relationship Management (CRM) system, and the data formatting module is further configured to generate visual sales funnels and marketing campaign analyses for sales and marketing departments. In yet another embodiment, the host application for embedding the marketplace is a Product Lifecycle Management (PLM) system, and the integration module is further configured to provide research and development data, including product specifications and test results formatted as technical documentation. In yet another embodiment, the host application for embedding the marketplace is a Governance, Risk, and Compliance (GRC) platform, and the access control module is further configured to enforce compliance with legal and regulatory standards by restricting access to compliance-related sensitive data. In one embodiment, the host application for embedding the marketplace is an IT service management tool, and the user interface module is further configured to display IT asset management data, system performance metrics, and security incident reports in a format suitable for use by the IT department. In another embodiment, the host application for embedding the marketplace is an enterprise intranet portal, and the data format module is further configured to provide management dashboards, departmental reports, and company-wide announcements in a centrally managed location. In yet another embodiment, the host application for embedding the marketplace is a cloud-based collaboration platform, and the integration module is further configured to facilitate data sharing and project management among geographically dispersed teams within the enterprise.

[0008] In embodiments, the technology described herein relates to a computer-implemented system for managing an embedded marketplace within an enterprise, the system comprising: a data classification module configured to classify enterprise data into classified data based on predefined confidentiality levels and regulatory compliance requirements; an access control module configured to manage permissions for different user roles within the enterprise and to grant access to the classified data in accordance with confidentiality levels and regulatory compliance requirements; a data formatting module configured to convert the classified data into formatted data in a display format customized for a set of enterprise departments; an integration module configured to interface with an enterprise system and to acquire and classify enterprise data, wherein the enterprise system includes at least one of an enterprise resource planning (ERP) system or a customer relationship management (CRM) system; and a user interface module configured to present the formatted data within a host application for accessing the embedded marketplace.

[0009] In an embodiment, the data classification module assigns data access permissions using role-based access control (RBAC). In an embodiment, the data classification module tags data with metadata indicating its sensitivity level. In an embodiment, the access control module includes the functionality of periodic auditing and real-time monitoring of data access. In an embodiment, the data formatting module provides management summaries with high-level graphics, dashboards, and overviews for executives. In an embodiment, the data formatting module provides detailed reports, raw datasets, and analytical tools for employees to perform detailed data analysis. In an embodiment, the user interface module enables customizable data views tailored to departmental needs. In an embodiment, the integration module includes a data services catalog with tools for processing, analyzing, and visualizing data. In an embodiment, the integration module is configured to extract relevant data from CRM, SCM, and PLM systems and format it for different departments. In an embodiment, the user interface module provides training modules and support services to assist employees in leveraging data. In an embodiment, the integration module provides a unified platform for centralized data governance across the enterprise. In an embodiment, the access control module supports attribute-based access control (ABAC) and RBAC. In embodiments, the integration module automates regulatory compliance by directly incorporating rules into the data access mechanism. In embodiments, the user interface module provides advanced search capabilities to improve data discoverability. In embodiments, the user interface module provides personalized data and service recommendations based on user roles and past usage. In embodiments, the integration module is configured to dynamically adjust permissions based on context, where that context includes at least one of the current project or collaboration.In embodiments, the integration module supports a scalable architecture capable of accommodating the growing volume and diversity of data. In embodiments, the user interface module provides an intuitive interface that eases the user's learning curve. In embodiments, the integration module includes usage tracking and analytics capabilities that provide insights into data value and usage patterns. In embodiments, the integration module maintains a comprehensive audit trail for security audits and compliance checks. In embodiments, the integration module facilitates subscription-based access to data services for predictable budgeting and cost management.

[0010] In this embodiment, the technology described herein relates to a computer-implemented method for integrating a system into a host platform for process automation and artificial intelligence, the method comprising: a step of identifying a set of functions provided by the host platform using a processing system; a step of determining a set of marketplace services associated with the identified functions of the host platform using a processing system; a step of integrating an interface for the marketplace services into the host platform, wherein the interface is configured to provide the marketplace services contextually based on user interaction with the host platform; a step of configuring the marketplace services using data from the host platform to personalize the marketplace services provided to the user using a processing system; and a step of facilitating transactions within the embedded marketplace without the user leaving the host platform using a processing system.

[0011] In an embodiment, the host platform includes an Enterprise Resource Planning (ERP) system, and the marketplace service is selected based on procurement needs identified by the ERP system. In an embodiment, the host platform includes a Customer Relationship Management (CRM) system, and the marketplace service is configured to provide products or services based on customer profiles and interactions stored within the CRM system. In an embodiment, the host platform includes a social media platform, and the marketplace service is configured to provide products or services related to content viewed by the user on the social media platform. In an embodiment, the host platform includes an Internet of Things (IoT) device, and the marketplace service is configured to provide maintenance, repair, or related products based on sensor data collected by the IoT device. In an embodiment, the host platform includes a digital wallet application, and the marketplace service is configured to provide financial products or services based on the user's financial transactions and preferences. In an embodiment, the host platform includes a content creation platform, and the marketplace service is configured to provide digital assets, tools, or services related to content being created by the user. In an embodiment, the host platform includes a gaming platform, and the marketplace service is configured to provide in-game items, virtual goods, or physical goods related to the game being played by the user. In an embodiment, the marketplace service includes an artificial intelligence algorithm for predicting user needs and proactively presenting relevant marketplace services within the host platform. In an embodiment, the marketplace service is configured to utilize process automation for processing transactions within the embedded marketplace, such transactions include settlement processing, order fulfillment, and post-transaction customer service.

[0012] In embodiments, the technology described herein further relates to a method comprising: a processing system collecting user feedback on a marketplace service; and a processing system using an artificial intelligence algorithm to adjust the marketplace service based on the collected feedback to improve relevance and user satisfaction within the embedded marketplace.

[0013] In embodiments, the technology described herein relates to a computer-implemented method for managing procurement within an enterprise, the method comprising: intercepting web browser communications initiated by enterprise employees; analyzing the intercepted communications to identify procurement-related activities; accessing regulatory databases to determine compliance with applicable laws and enterprise policies; evaluating procurement requests based on budget constraints and employee approval levels; and controlling the execution of procurement transactions by allowing, modifying, or blocking them based on compliance and approval assessments.

[0014] In embodiments, the technologies described herein relate to a system for automated procurement management in an enterprise environment and include: a network communications analytics module configured to monitor and evaluate web-based procurement activities; a compliance evaluation engine integrated with a regulatory database for real-time compliance verification; an approval management module for facilitating and tracking the approval process of procurement requests; and a transaction execution module for enforcing compliance and approval results by managing the finalization of procurement transactions.

[0015] In embodiments, the technology described herein relates to a computer-implemented system for integrating a marketplace into a digital twin, the system comprising: a processing system configured to generate a digital twin representing a physical asset, the processing system comprising real-time data in which the digital twin reflects the state, status, and performance of the physical asset; a marketplace module embedded within the digital twin, configured to facilitate transactions related to the physical asset, the marketplace module comprising listing, purchasing, and transaction processing functions; a data analysis module configured to use real-time data from the digital twin to identify transaction needs or opportunities within the marketplace module; and a communication interface that presents transaction opportunities to a user and enables the user to interact with the marketplace module via the digital twin.

[0016] In embodiments, the marketplace module is further configured to provide predictive maintenance services for physical assets based on analysis, and further configured to provide recommendations for spare parts and consumables compatible with the physical assets. In embodiments, the marketplace module includes smart contract functionality configured to automate the execution of transactions based on predetermined rules derived from real-time data. In embodiments, the marketplace module is further configured to provide insurance services, where the terms of the insurance services are dynamically adjusted based on real-time data from the digital twin. In embodiments, the marketplace module is further configured to facilitate the resale or lease of physical assets by connecting potential buyers or lessees with the digital twin. In embodiments, the marketplace module is further configured to collaborate with third-party service providers to enable the provision of extended services related to physical assets. In embodiments, the marketplace module is further configured to utilize machine learning algorithms to personalize transaction opportunities presented to the user based on the user's behavior and preferences.

[0017] In embodiments, the marketplace module is further configured to support a virtual reality interface, enabling users to interact with the digital twin and the marketplace in an immersive environment. In embodiments, the marketplace module is further configured to provide a platform for user-generated content, allowing users to post custom modifications and improvements related to physical assets. In embodiments, the marketplace module is further configured to aggregate data from multiple digital twins representing a fleet, and the marketplace module is further configured to enable energy trading services based on real-time energy usage and production data for digital twins representing energy-consuming or energy-generating assets. In embodiments, the marketplace module is further configured to provide subscription-based services related to physical assets, with subscription terms that can be modified in response to changes in real-time data. In embodiments, the marketplace module is further configured to provide a feedback mechanism for users to evaluate and review transactions, which influences the presentation of future transaction opportunities within the marketplace. In embodiments, the marketplace module is further configured to support regulatory compliance monitoring, with transactions automatically adjusted to comply with applicable laws and regulations based on real-time data.

[0018] In embodiments, the technology described herein relates to a system that provides an integrated transaction platform, the system comprising: an embedded marketplace module configured to aggregate offerings from multiple vendors within a user interface of a host application; a data aggregation system configured to collect and process data from various sources and to personalize the aggregated offerings based on user preferences and behavior; a transaction execution module configured to facilitate the purchase, sale, and exchange of goods and services within the embedded marketplace; a blockchain interface configured to work with one or more distributed ledgers to record transactions performed within the embedded marketplace; and a smart contract module configured to generate and execute contracts related to transactions within the embedded marketplace based on predetermined rules and conditions.

[0019] In embodiments, the embedded marketplace module is further configured to present a unified view of aggregated offerings across multiple external marketplaces. In embodiments, the data aggregation system utilizes machine learning algorithms to refine personalization based on real-time user interaction with the embedded marketplace. In embodiments, the transaction execution module is further configured to process payments using at least one of fiat currency or cryptocurrency. In embodiments, the blockchain interface is further configured to support multiple blockchain protocols to ensure compatibility with various distributed ledger technologies. In embodiments, the smart contract module is further configured to automatically adjust contract terms based on changes in regulatory requirements. In embodiments, the technology described herein relates to a system further comprising a robotic process automation (RPA) module configured to automate procurement processes based on inventory levels and forecast demand analysis. In embodiments, the RPA module is further configured to work with a vendor management system to streamline supply chain operations. In embodiments, the embedded marketplace module is further configured to work with a digital twin representation of physical assets to enhance the visualization of offerings. In embodiments, the transaction execution module includes a recommendation engine that suggests ancillary services related to key offerings. In an embodiment, the blockchain interface is configured to tokenize assets in order to facilitate asset trading within the embedded marketplace. In an embodiment, the smart contract module includes a dispute resolution mechanism that automatically activates based on transaction anomalies. In an embodiment, the data aggregation system is further configured to aggregate at least one of social media data and IoT device data in order to enhance the personalization of offerings. In an embodiment, the embedded marketplace module is further configured to provide location-based services and to provide goods and services related to the user's geographical location.In an embodiment, the transaction execution module is further configured to support subscription-based transactions corresponding to recurring purchases within the embedded marketplace. In an embodiment, the blockchain interface is further configured to provide an audit trail of transactions to ensure transparency and compliance. In an embodiment, the smart contract module is further configured to work with an external contract management system for contract synchronization across platforms. In an embodiment, the RPA module is further configured to automate compliance checks against corporate policies during the procurement process. In an embodiment, the embedded marketplace module is further configured to embed the marketplace within a virtual reality environment to deliver an immersive shopping experience. In an embodiment, the transaction execution module is further configured to leverage the blockchain interface and smart contract module to enable peer-to-peer transactions without the involvement of intermediaries.

[0020] In this embodiment, the technology described herein relates to a computer-implemented system for managing transactions within an enterprise ecosystem, the system including a processor and memory that stores instructions, when executed by the processor, causing the system to perform the following actions: integrating an enterprise access layer (EAL) that interfaces with multiple enterprise resources with an embedded marketplace; automating procurement and sales processes by interface the embedded marketplace with the enterprise's workflow system; utilizing a data services system for managing listing information, transactions, and user profiles within the embedded marketplace; implementing an intelligence system that provides predictive analytics for market trends and demand forecasting within the embedded marketplace; enforcing security and compliance through an authorization system that controls access to the functions of the embedded marketplace; managing digital transactions through a wallet system that interfaces with the embedded marketplace; and generating reports on marketplace activities through a reporting system that communicates with the embedded marketplace.

[0021] In an embodiment, the command further causes the system to collect real-time data, analyze real-time data, and provide personalized recommendations for products and services based on the user's past transaction data. In an embodiment, the command further causes the system to implement a smart contract orchestration engine and automate transaction workflows within the enterprise ecosystem. In an embodiment, the command further causes the system to work in conjunction with technologies deployed on the enterprise's private network, the private network including at least one of on-premises resources and platforms or cloud resources and platforms. In an embodiment, the command further causes the system to tokenize digital assets and digitally represent transactions within the enterprise ecosystem. In an embodiment, the command further causes the system to facilitate transactions with external entities by providing a set of network resources for bilateral or multilateral transactions involving the enterprise. In an embodiment, the command further causes the system to simplify transactions for the enterprise by enabling the enterprise to interact with multiple markets, marketplaces, exchanges, and platforms through a common access point. In an embodiment, the instruction further causes the system to employ blockchain to manage and protect transactions within the enterprise ecosystem. In an embodiment, the instruction further causes the system to include a generative content system that utilizes a large-scale language model (LLM) trained on enterprise-specific data to propose new workflows for enterprise processes.

[0022] In an embodiment, the technology described herein relates to a computer-implemented system for facilitating transactions within an embedded marketplace enterprise ecosystem, the system including a processor and a memory storing instructions that, when executed by the processor, cause the system to: integrate an embedded marketplace with an enterprise's digital infrastructure; automate the transaction process by interfacing the embedded marketplace with an enterprise's workflow system; manage listing information, transactions, and user profiles within the embedded marketplace using a data service system; provide analysis and insights for strategic decision-making within the embedded marketplace through an intelligence system; enforce security and compliance protocols via an authorization system that controls access to the embedded marketplace; and facilitate digital transactions through a wallet system that interfaces with the embedded marketplace.

[0023] In embodiments, the embedded marketplace leverages a large-scale language model (LLM) trained on enterprise-specific data to assist in the generation and optimization of transaction workflows. In embodiments, the embedded marketplace employs robotic process automation (RPA) to streamline procurement and sales processes by automating repetitive tasks and data processing. In embodiments, the embedded marketplace includes a digital twin of the enterprise ecosystem that simulates and analyzes marketplace dynamics and enterprise resource planning scenarios. In embodiments, the embedded marketplace works in conjunction with a blockchain network to manage and protect transactions and ensure data integrity and traceability. In embodiments, the embedded marketplace is configured to use artificial intelligence (AI) for dynamic pricing strategies based on real-time market data and predictive analytics. In embodiments, the embedded marketplace incorporates a smart contract orchestration engine to automate agreement execution and compliance with contract terms. In embodiments, the embedded marketplace is interfaceable with Internet of Things (IoT) devices to facilitate transactions based on sensor data and automated triggers. In embodiments, the embedded marketplace utilizes machine learning algorithms to personalize product recommendations based on user behavior and preferences. In embodiments, the embedded marketplace is further configured to support subscription services, enabling recurring transactions and customer retention strategies. In embodiments, the embedded marketplace features a virtual assistant leveraging natural language processing (NLP) to help users navigate the marketplace and complete transactions. In embodiments, the embedded marketplace is designed to work in conjunction with virtual reality and augmented reality (VR / AR) platforms, providing immersive product demonstrations and virtual showrooms.In embodiments, the embedded marketplace is configured to tokenize digital assets representing ownership and transactions of digital and physical goods within the enterprise ecosystem. In embodiments, the embedded marketplace is further configured to facilitate cross-border transactions by incorporating multi-currency and multi-language support. In embodiments, the embedded marketplace enhances customer service and engagement by employing a customer relationship management (CRM) system to track customer interactions and transactions. In embodiments, the embedded marketplace is further configured to optimize inventory levels and logistics in conjunction with a supply chain management system. In embodiments, the embedded marketplace includes an API gateway that allows third-party applications and services to interact with the marketplace ecosystem. In embodiments, the embedded marketplace is further configured to employ a fraud detection system that identifies and prevents fraudulent transactions using anomaly detection technology. In embodiments, the embedded marketplace is configured to support a peer-to-peer (P2P) network for direct transactions between users without intermediary involvement. In embodiments, the embedded marketplace includes a feedback and evaluation system that employs sentiment analysis to measure customer satisfaction and improve service delivery.

[0024] The characteristics of these and other features, the properties of this technology, and the operation and function of the related structural elements, the combination of parts, and the economy of the manufacturer will become clearer by considering the following description and the attached claims with reference to the attached drawings, all of which constitute part of this specification, and the same reference numerals indicate the corresponding parts in each drawing. However, it should be explicitly understood that the drawings are for illustrative and illustrative purposes only and do not define the limits of the invention. The singular forms "a," "an," and "the" used in the specification and claims refer to plural subjects unless the context clearly indicates otherwise. A more complete understanding of the disclosure will be obtained from the following description, attached drawings, and claims.

Brief Description of the Drawings

[0025] The following detailed description of the present disclosure and its specific embodiments can be understood by referring to the following figures.

[0026] [Figure 1] Figure 1 is a schematic diagram of the components of a platform that enables intelligent transactions according to an embodiment of the present disclosure.

[0027] [Figure 2A] Figures 2A and 2B are schematic diagrams of additional components of a platform that enables intelligent transactions according to an embodiment of the present disclosure. [Figure 2B] Figures 2A and 2B are schematic diagrams of additional components of a platform that enables intelligent transactions according to an embodiment of the present disclosure.

[0028] (Diagram of the Intelligence Service System) [Figure 3] Figure 3 is a schematic diagram showing an example of an intelligent service system according to some embodiments.

[0029] [Figure 4] Figure 4 is a schematic diagram showing an example of a neural network according to some embodiments.

[0030] [Figure 5] Figure 5 is a schematic diagram of an example of a convolutional neural network according to some embodiments.

[0031] [Figure 6] Figure 6 is a schematic diagram of an example of a neural network according to some embodiments.

[0032] [Figure 7] Figure 7 is an illustration of an approach based on reinforcement learning according to some embodiments.

[0033] [Figure 8] Figure 8 shows a block diagram illustrating the exemplary features, capabilities, and interfaces of a robust generative artificial intelligence platform.

[0034] (Diagram of the enterprise access layer) [Figure 9] Figure 9 is a schematic diagram illustrating an example of an enterprise ecosystem, including the enterprise access layer.

[0035] [Figure 10] Figure 10 is a functional block diagram showing an example implementation of the enterprise access layer.

[0036] [Figure 11] Figure 11 is a schematic diagram illustrating an example of how the enterprise access layer in Figure 10 is integrated with parts of the enterprise ecosystem.

[0037] [Figure 12] Figure 12 is a schematic diagram illustrating an example of a market orchestration system including an enterprise access layer.

[0038] [Figure 13] Figure 13 is a functional block diagram of an exemplary implementation of an intelligence system.

[0039] [Figure 14] Figure 14 is a functional block diagram of an exemplary implementation of a data pool system.

[0040] [Figure 15] Figure 15 is a functional block diagram of an exemplary implementation of the scoring system.

[0041] [Figure 16]Figure 16 shows a simplified diagram of how attention level is determined by a machine learning model according to some of the examples.

[0042] [Figure 17] Figure 17 is a simplified diagram of a transformer model according to several embodiments.

[0043] [Figure 18] Figure 18 is a simplified diagram of a financial infrastructure system according to several embodiments.

[0044] (Diagram of process automation and artificial intelligence) [Figure 19] Figure 19 shows an exemplary block diagram of a trading environment (e.g., a marketplace or a set of marketplaces) according to an exemplary embodiment of the present disclosure.

[0045] [Figure 20] Figure 20 shows an exemplary block diagram of a system that implements a processing system for automating market transactions, according to an exemplary embodiment of the disclosure.

[0046] [Figure 21] Figure 21 shows an exemplary block diagram illustrating the processing system of Figure 20, according to an embodiment of the present disclosure, and showing various modules within it.

[0047] [Figure 22] Figure 22 shows an exemplary flowchart for automating market transactions according to an embodiment of the present disclosure.

[0048] [Figure 23] Figure 23 shows an exemplary block diagram of a system that implements a processing system for managing transactions in the market, according to an embodiment of the present disclosure.

[0049] [Figure 24]Figure 24 shows an exemplary block diagram illustrating various modules of the processing system shown in Figure 23, in accordance with an exemplary embodiment of the disclosure.

[0050] [Figure 25] Figure 25 provides an exemplary flowchart for automating market transactions according to an embodiment of the present disclosure.

[0051] [Figure 26] Figure 26 shows an exemplary block diagram of a system that implements a processing system for automating the processing of market transactions, according to an embodiment of the present disclosure.

[0052] [Figure 27] Figure 27 shows an exemplary block diagram illustrating various modules of the processing system shown in Figure 26, in accordance with an exemplary embodiment of the disclosure.

[0053] [Figure 28] Figure 28 provides an exemplary flowchart for automating market transaction processing according to an embodiment of the present disclosure.

[0054] [Figure 29] Figure 29 shows an exemplary block diagram of a system for automated orchestration in the market, according to an embodiment of the present disclosure.

[0055] [Figure 30] Figure 30 shows an exemplary flowchart for automated orchestration in the market according to an embodiment of the present disclosure.

[0056] [Figure 31] Figure 31 shows an exemplary block diagram of a system for service expansion in the market, according to an embodiment of the present disclosure.

[0057] [Figure 32]Figure 32 shows an exemplary flowchart for service expansion in the market according to an embodiment of the present disclosure.

[0058] [Figure 33] Figure 33 is a schematic diagram of an embedded marketplace system according to an embodiment of the present disclosure.

[0059] [Figure 34] Figure 34 is a schematic diagram of an embedded marketplace platform used in an embedded marketplace system according to an embodiment of the present disclosure. [Modes for carrying out the invention]

[0060] (Trading platform) Referring to Figures 1, 2A, and 2B, a series of systems, methods, components, modules, machines, articles, blocks, circuits, services, programs, applications, hardware, software, and other elements are provided, which are collectively referred to herein as System or Platform 100. Platform 100 enables a wide range of improvements for various machines, systems, and other components that enable trading involving the exchange of value (including currencies, cryptocurrencies, tokens, rewards, etc., and various physical and other resources) relating to various goods, services, and resources in various markets (including spot markets 170, futures markets 130, etc.). In this specification, “currency” should be understood to include legal tender issued or regulated by governments, cryptocurrencies, value tokens, tickets, loyalty points, reward points, coupons, and other elements that represent or are exchangeable for value. Resources exchangeable for value in the market include goods, services, natural resources, energy resources, computing resources, energy storage resources, data storage resources, network bandwidth resources, processing resources, etc. This includes both the resources whose value is exchanged and the resources that enable the trading (such as computing and processing resources, storage resources, network resources, and energy resources necessary to enable the trading). Platform 100 may include a set of futures trading machines 110. Each machine may be configured as an expert system or an automated intelligent agent to interact with one or more of the set of spot markets 170 and futures markets 130.The following systems enable the futures trading machine group 110: an intelligent resource purchasing system 164 with a group of intelligent agents for purchasing resources in the spot and futures markets; an intelligent resource allocation and adjustment system 168 for the intelligent sale of allocated or adjusted resources (such as computing resources, energy resources, and other resources involved in or enabling trading); an intelligent sales engine 172 for intelligently adjusting the sale of allocated resources in the spot and futures markets; and an automated spot market testing and arbitrage trading execution engine 194 that performs spot testing of the spot and futures markets using micro-trades, etc., and automatically executes resource trades that take advantage of favorable arbitrage conditions when they are present. Each engine can utilize model-based or rule-based expert systems based on rules and heuristics, as well as deep learning systems that learn rules and heuristics through trials with large input sets. The engines can utilize any expert system and artificial intelligence capabilities described throughout this specification. Interactions within platform 100 (including interactions between all platform components, between them, and with various markets) are tracked and collected by data aggregation systems 144, etc. For example, this is to aggregate buying and selling data from various marketplaces by the machines described herein. Aggregated data may include tracking and result data that can be supplied to artificial intelligence and machine learning systems (for training or supervising them, etc.). The various engines operate on a variety of data sources. These include aggregated data from marketplace transactions, tracking data on the operation of each engine, and a group of external data sources 182. The group of external data sources 182 includes social media data sources 180 (social networking sites such as Facebook® and Twitter®), and Internet of Things (IoT) data sources (including data from sensors, cameras, data acquisition devices, and measuring instruments and systems).For example, IoT sources that provide information on equipment and systems that enable transactions, or equipment and systems that are involved in the production and consumption of resources. External data sources 182 include automated agent behavior data sources 188 (such as tracking and reporting the behavior of automated agents for conversation and dialogue management, machine and system control functions, purchasing and selling, data collection, advertising, etc.), human behavior data sources (such as data sources that track online behavior, mobility behavior, energy consumption behavior, energy production behavior, network usage behavior, computational processing behavior, resource consumption behavior, resource production behavior, purchasing behavior, attention behavior, social behavior, etc.), and entity behavior data sources 190 (including the behavior of business organizations and other entities, such as purchasing behavior, consumption behavior, production behavior, market activities, merger and acquisition behavior, transaction behavior, location-based behavior, etc.). IoT, social, and behavioral data from or relating to sensors, machines, humans, entities, and automated agents may be used collectively to build expert systems, machine learning systems, and other intelligent systems and engines as described throughout this disclosure. For example, data may be provided as input to a deep learning system and as feedback or results for the purpose of training, supervising, and iteratively improving the system for prediction, forecasting, classification, automation, and control. Data may be structured as an event stream. Data may be stored in a distributed ledger or other distributed system. Data may be stored in a knowledge graph where nodes represent entities and links represent relationships. External data sources can be queried through various database query functions. External data sources are accessible through APIs, brokers, connectors, protocols such as REST and SOAP, and other data ingestion and extraction techniques. Data may be enhanced with metadata and subject to transformation and loading into a format suitable for use by the engine (e.g., cleansing, normalization, deduplication, etc.).

[0061] Platform 100 may include a set of intelligent predictive engines 192 for predicting events, activities, variables, and parameters such as a spot market 170, a futures market 130, the resources traded in those markets, the resources that enable those markets, behavior (including those tracked by an external data source 182), and trades. The intelligent predictive engines 192 operate on data from a data aggregation system 144 regarding the components of Platform 100 and data from the external data source 182. The platform may include a set of intelligent trading engines 136 for automatically executing trades in the spot market 170 and the futures market 130. This includes executing intelligent cryptocurrency trades using an intelligent cryptocurrency execution engine 183 associated with IoT data 295 for cryptocurrency trading and social data 193 for cryptocurrency trading. Platform 100 may leverage the assets of an improved distributed ledger 113 and improved smart contracts 103. This includes incorporating and operating proprietary information, instruction sets, etc., to enable complex peer-to-peer transactions with reduced (or eliminated) reliance on intermediaries. These and other components are described in more detail throughout this disclosure.

[0062] Referring to the block diagrams in Figures 2A and 2B, details of platform 100 and additional components, as well as their interactions, are shown. The futures trading machine group 110 may include a renewable capacity allocation engine 102 (for example, for allocating energy generation or regeneration capacity; for example, when allocating energy to one or more of the following purposes within a system including a hybrid vehicle or energy generation / regeneration capacity, a renewable energy system with energy storage capabilities, or other energy storage systems: sale in the futures market 130, sale in the spot market 170, use for trade completion (e.g., cryptocurrency mining), or other purposes). For example, the renewable capacity allocation engine 102 may consider available options for the use of stored energy, such as sale in current and futures energy markets that accept energy from producers, storage and maintenance of energy for future use, or use of energy for operations (including processing activities such as platform processing activities such as data collection and processing, and processing operations such as processing operations for trade execution (including cryptocurrency mining activities)). In embodiments, energy storage capacity may be traded in the energy storage futures market 174 or the energy storage market 178.

[0063] The futures trading equipment group 110 may include an energy trading device 104 for buying and selling energy in the energy spot market 148 and the energy futures market 122, etc. The energy trading machine 104 may use expert systems, neural networks, and other intelligences to determine the timing of purchases based on current and predicted status information regarding energy prices and supply conditions, as well as current and predicted status information regarding energy demand (such as computing tasks, cryptocurrency mining, data collection activities, and work by automated agents and systems, or work based on human behavior or the behavior of entities). For example, the energy purchasing device may use machine learning to recognize the possibility that a business operator will request energy blocks necessary for increased production in manufacturing based on increased orders or expanding market demand, and then execute an energy purchase at a favorable price in the futures market based on a combination of energy market data and real-world behavior data. Continuing this example, market demand may be understood by machine learning, for example, by processing human behavior data sources 184 such as social media posts and e-commerce data indicating increased demand. The energy trading machine 104 sells energy in the spot energy market 148 or the energy futures market 122. Sales are also carried out by expert systems operating on various data sources described herein, including training and human supervision regarding the results.

[0064] The futures trading system group 110 may include a renewable energy credit (REC) trading system 108. This system purchases renewable energy credits, pollution credits, and other environmental and regulatory credits in the spot market 150 or futures market 124 for the applicable credits. Purchases are configured and managed by expert systems operating on data aggregated by one of the external data sources 182 or by the platform data aggregation system group 144. Renewable energy credits and other credits are purchased by automated systems using expert systems (including machine learning and other artificial intelligence). For example, purchases may be made at favorable times based on an understanding of supply and demand determined by processing input from data sources. Expert systems are trained using datasets of purchase results under historical input conditions. They may also be trained using datasets of human purchase decisions or operated under the supervision of one or more operators. The renewable energy credit (REC) trading machine 108 can also sell renewable energy credits, pollution credits, and other environmental and regulatory credits in the spot market 150 or futures market 124. Sales are also carried out by expert systems operating on the various data sources described herein, including learning from results and human supervision.

[0065] The futures trading machine group 110 may include a attention resource trading machine 112, which purchases one or more attention-related resources in the attention spot market 152 or the attention futures market 128, such as advertising space, search lists, keyword lists, banner ads, participation in panel / survey activities, and participation in trial / pilot activities. Attention resources may include the attention of automated agents such as bots, crawlers, and conversation managers used for searching, shopping, and purchasing. The purchase of attention resources may be configured and managed by expert systems operating on data aggregated by one of the external data sources 182 for the platform or by data aggregated by the data aggregation system group 144. Attention resources may be purchased by automated systems that utilize expert systems, including machine learning and other artificial intelligence. For example, resource purchases may be made at favorable times based on an understanding of supply and demand determined by processing inputs from various data sources. For example, the attention buy / sell machine 112 can purchase advertising space in the advertising futures market based on what it has learned from a wide range of inputs, such as market conditions, behavioral data, and data on the activities of agents and systems within the platform 100. Expert systems may be trained using purchase result datasets under historical input conditions. The expert system may be trained using human purchase decision datasets or operated under the supervision of one or more operators. The attention trading machine 112 may also sell one or more attention-related resources, such as advertising space, search lists, keyword lists, banner ads, participation in panel or survey activities, participation in trials or pilots, or similar, in the attention spot market 152 or the attention futures market 128. This may include access to one or more automated agents of platform 100, or the provision or sale of attention to them. Sales may also be carried out by expert systems operating on the various data sources described herein, including training on results and human supervision.

[0066] The futures trading machine group 110 may include a compute buy / sell machine 114, which purchases one or more compute-related resources such as processing resources, database resources, compute resources, server resources, disk resources, I / O resources, temporary storage resources, memory resources, virtual machine resources, and container resources in the spot market 154 or the futures market 132. The purchase of compute resources may be configured and managed by an expert system operating on data aggregated by one of the external data sources 182 for the platform or by a data aggregation system group 144. Computing resources may be purchased by an automated system that utilizes an expert system, including machine learning and other artificial intelligence. For example, this might involve purchasing resources at a favorable time (e.g., during periods of low demand) based on an understanding of supply and demand determined by processing inputs from various data sources. For example, the compute buy / sell machine 114 purchases or reserves compute resources on the cloud platform in the futures market for compute resources based on what it has learned from a wide range of inputs, such as market conditions, behavioral data, and data on the activities of agents and systems within platform 100. This makes it possible to acquire resources at favorable prices during periods of surges in computing demand. The expert system is trained using a dataset of purchase results under past input conditions. The expert system may be trained using a dataset of human purchase decisions and / or may be supervised by one or more human operators. The compute purchase and sales machine 114 may also sell one or more compute-related resources, such as processing resources, database resources, compute resources, server resources, disk resources, I / O resources, temporary storage resources, memory resources, virtual machine resources, and container resources, which are connected to, part of, or managed by platform 100, on the spot market 154 or the futures market 132.Sales are also conducted by expert systems operating on various data sources described herein, including training on results and human supervision.

[0067] The futures trading machine group 110 may include a data storage trading machine 118, which purchases database resources, disk resources, server resources, memory resources, RAM resources, network-attached storage resources, storage-attached network (SAN) resources, tape resources, time-based data access resources, virtual machine resources, container resources, etc., in the spot market for storage resources 158 or the futures market for data storage 134. The purchase of data storage resources may be configured and managed by an expert system operating on data aggregated by one of the external data sources 182 or by the platform data aggregation system group 144. Data storage resources may be purchased by an automated system that utilizes an expert system, including machine learning and other artificial intelligence. For example, resources may be purchased at a favorable time (e.g., based on an understanding of supply and demand) based on an understanding of supply and demand determined by processing inputs from various data sources. For example, the compute buy / sell machine 114 purchases or reserves computing resources on the cloud platform in the futures market for computing resources based on results learned from a wide range of inputs, such as market conditions, behavioral data, and data on the activities of agents and systems within platform 100. This makes it possible to acquire resources at favorable prices during periods of surge in storage demand. Expert systems are trained using datasets of purchase results under historical input conditions. Expert systems may be trained using datasets of human purchase decisions or operated under the supervision of one or more operators. The data storage trading machine 118 can also sell one or more data storage-related resources connected to, forming part of, or managed by platform 100 in the spot market 158 ​​for storage resources or the futures market 134 for data storage. Sales are also carried out by expert systems operating on various data sources described herein, including training on results and human supervision.

[0068] The futures trading device group 110 may include a bandwidth trading device 120, which purchases one or more bandwidth-related resources (e.g., cellular bandwidth, Wi-Fi bandwidth, wireless bandwidth, access point bandwidth, beacon bandwidth, local area network bandwidth, wide area network bandwidth, enterprise network bandwidth, server bandwidth, storage I / O bandwidth, advertising network bandwidth, market bandwidth, and other bandwidth) in the spot market 160 for bandwidth resources or the futures market 138 for bandwidth resources. The purchase of bandwidth resources may be configured and managed by an expert system operating on data aggregated by one of the external data sources 182 or by a group of data aggregation systems 144 for the platform. Bandwidth resources may be purchased by an automated system using an expert system, including machine learning and other artificial intelligence. For example, resource purchases may be made at favorable times based on an understanding of supply and demand determined by processing inputs from various data sources. For example, the bandwidth trading machine 120 buys or reserves bandwidth on network resources for future network activity based on learning from a wide range of inputs, including market conditions, behavioral data, and data on the activities of agents and systems within platform 100. This makes it possible to acquire resources at favorable prices during periods of surge in bandwidth demand. Expert systems are trained using datasets of purchase results under historical input conditions. Expert systems may be trained using datasets of human purchase decisions or may operate under the supervision of one or more operators. The bandwidth trading machine 120 can also sell one or more bandwidth-related resources connected to, constituting part of, or managed by platform 100 in the spot market 160 or futures market 138 for bandwidth resources. Sales are also carried out by expert systems operating on various data sources described herein, including learning from results and human supervision.

[0069] The futures trading equipment group 110 may include a spectrum trading equipment 142, which purchases spectrum resources such as cellular spectrum, 3G spectrum, 4G spectrum, LTE spectrum, 5G spectrum, cognitive radio spectrum, peer-to-peer network spectrum, and emergency responder spectrum in the spot market 162 for spectrum resources or the futures market 140 for spectrum / bandwidth. The purchase of spectrum resources may be configured and managed by an expert system operating on data aggregated by one of the external data sources 182 for the platform or by a data aggregation system group 144. Spectrum resources may be purchased by an automated system using expert systems, including machine learning and other artificial intelligence. For example, resource purchases may be made at favorable times based on an understanding of supply and demand determined by processing inputs from various data sources. For example, the spectrum trading machine 142 purchases or reserves spectrum on network resources for future network activity based on results learned from a wide range of inputs, such as market conditions, behavioral data, and data on the activities of agents and systems within platform 100. This makes it possible to acquire resources at favorable prices when spectrum demand surges. The expert system is trained using a dataset of purchase results under historical input conditions. The expert system may be trained using a dataset of human purchase decisions or may operate under the supervision of one or more operators. The spectrum buy / sell machine 142 may also sell one or more spectrum-related resources connected to, forming part of, or managed by the platform 100 in the spectrum resource spot market 162 or the spectrum / bandwidth futures market 140. Sales are also carried out by expert systems operating on various data sources described herein, including training on results and human supervision.

[0070] In embodiments, an intelligent resource allocation and adjustment system 168, including an intelligent resource purchasing system 164, an intelligent sales engine 172, and an automated spot market testing and arbitrage trading execution engine 194, can provide coordinated automated allocation of resources and coordinated execution of trades across various futures markets 130 and spot markets 170 by coordinating various purchase and sales machines, for example, by expert systems such as machine learning systems (which may be model-based or deep learning systems, trained on results, and / or supervised by humans). For example, the intelligent resource allocation and coordination system 168 may include asset groups (vehicle groups, data centers with processing and data storage resources, information technology networks (on-premise, cloud, or hybrid), energy production systems (renewable / non-renewable), smart homes / buildings (including appliances, machinery, infrastructure components / systems that consume / produce resources). The platform 100 can optimize the allocation of resource purchases, sales, and uses based on aggregated data within the platform, such as by tracking the activities of various engines and agents and incorporating inputs from external data sources 182. In embodiments, the outcomes (yield, profitability, resource optimization, etc.) may be considered. Based on factors such as optimization of business objectives, degree of objective achievement, and user or operator satisfaction, etc., this may be provided as training feedback for the intelligent resource allocation and coordination system 168. For example, as the energy consumption of computing tasks comes to account for a significant portion of a company's energy use, platform 100 can learn how a set of machines with energy storage capacity can optimally allocate that capacity among computing tasks (such as cryptocurrency mining, neural network application, and computation on data), other useful tasks (which may yield profits or other benefits), storage for future use, or sales to energy grid providers.Platform 100 could be used by fleet operators, corporations, governments, local authorities, military forces, emergency response forces, manufacturers, energy producers, cloud platform providers, and other companies and operators that own or operate resources that consume or provide energy, computing, data storage, bandwidth, and spectrum. Platform 100 can also be used in connection with attention markets, such as advertising markets and microtransaction markets, to support attention-based value exchange by leveraging available resource capacity.

[0071] Referring to Figures 2A and 2B, platform 100 may include a set of intelligent predictive engines 192. These predict one or more attributes, parameters, variables, or other factors and are used, for example, as input to a set of futures trading machines, an intelligent trading engine 136 (such as intelligent cryptocurrency execution), or for other purposes. Each of the intelligent predictive engines 192 can use data tracked, aggregated, processed, or handled within platform 100 by a data aggregation system 144, etc., as well as input data from external data sources 182 (such as input data obtained from social media data source 180, automated agent behavior data source 188, human behavior data source 184, entity behavior data source 190, IoT data source 198, etc.). These collective inputs may be used to predict attributes using models (e.g., Bayesian, regression, or other statistical models), rules, or expert systems (e.g., a machine learning system having one or more classifiers, pattern recognizers, and predictors, such as any of the expert systems described throughout this specification). In embodiments, the intelligent predictive engines 192 may include one or more specialized engines that predict market attributes such as capacity, demand, supply, and price using specific data sources for a particular market. These may include an energy price prediction engine 215 that makes predictions based on the behavior of automated agents, a network spectrum price prediction engine 217 that makes predictions based on the behavior of automated agents, a renewable energy certificate (REC) price prediction engine 219 that makes predictions based on the behavior of automated agents, a computing price prediction engine 221 that makes predictions based on the behavior of automated agents, and a network spectrum price prediction engine 223 that makes predictions based on the behavior of automated agents. In any case, observations about the behavior of automated agents used for conversation, dialogue management, e-commerce management, advertising management, etc., may be provided as input to the prediction engines.The intelligent prediction engine 192 may also include a set of engines that provide predictions based at least in part on the actions of entities (such as the actions of companies and other organizations, including marketing actions, sales actions, product offering actions, advertising actions, purchasing actions, transaction actions, merger and acquisition actions, and other entity actions). These include an energy price prediction engine 225 that utilizes entity actions, a network spectrum price prediction engine 227 that utilizes entity actions, a renewable energy certificate (REC) price prediction engine 229 that utilizes entity actions, a computing price prediction engine 231 that utilizes entity actions, and a network spectrum price prediction engine 233 that utilizes entity actions.

[0072] The intelligent predictive engine 192 may also include a set of engines that provide predictions based at least in part on human behavior, such as consumer behavior and user behavior (including purchasing behavior, shopping behavior, selling behavior, product interaction behavior, energy use behavior, mobility behavior, activity level behavior, activity type behavior, transaction behavior, and other human behavior). These include a human behavior-based energy price prediction engine 235, a human behavior-based network spectrum price prediction engine 237, a human behavior-based renewable energy certificate (REC) price prediction engine 239, a human behavior-based computing price prediction engine 241, and a human behavior-based network spectrum price prediction engine 243.

[0073] Referring to Figures 2A and 2B, platform 100 may include a set of intelligent trading engines 136 that automate the execution of trades in the futures market 130 and / or the spot market 170 when favorable conditions are determined to exist. This determination is made by an intelligent resource allocation and coordination system 168 and / or by utilizing predictions from an intelligent forecasting engine 192. The intelligent trading engines 136 may be configured to automatically execute trades in each of the above markets using available market interfaces such as APIs, connectors, ports, and network interfaces. In embodiments, the intelligent trading engines may execute trades based on event streams obtained from external data sources such as an IoT data source 198 and a social media data source 180. The engines may include, for example, an IoT forward energy trading engine 195 and / or an IoT computing market trading engine 106, either or both of which may use data from the Internet of Things to determine the timing and other attributes of market trades in the markets of one or more resources described herein (such as energy market trading, computing resource trading, and other resource trading). IoT data includes measurement and control data from one or more machines (arranged as a fleet) that use or produce energy or use or own computing resources; weather data that affects energy prices and consumption (such as wind data that affects wind energy production); sensor data from energy production environments; sensor data from points of use of energy or computing resources (such as vehicle traffic data, network traffic data, IT network utilization data, internet utilization and traffic data, camera data from work sites, smart building data, smart home data, etc.); and other data collected or transmitted by the Internet of Things (IoT), including data stored on IoT platforms and cloud service providers such as Amazon and IBM.The intelligent trading engine 136 may include an engine that uses social data to determine the timing and other attributes of market trades in one or more resources described herein. Examples include the social data futures energy trading engine 199 and / or the social data computing market trading engine 116. Social data includes data from social networking sites (e.g., Facebook®, YouTube®, Twitter®, Snapchat®, Instagram®, etc.), websites, e-commerce sites, and other sites containing information that may be relevant to determining or predicting user or entity behavior (such as data indicating interest in or attention to specific topics, goods, or services; and data indicating the type and level of activities in which an individual is engaged in activities, such as travel, work, or leisure activities, which may be observed by machine processing as image data). Social data may be supplied to machine learning, such as learning user or entity behavior in the social data market forecaster 186. It may also be supplied as input to specialized systems, models, etc., that determine trading parameters based on social data. For example, an event or series of events within a social data stream may indicate a potential surge in interest in an online resource, product, or service, and it may be possible to pre-purchase computing resources, bandwidth, storage, etc., to respond to the increased interest reflected in the social data stream (avoiding price spikes during the surge).

[0074] (Neural network system) Embodiments of this disclosure, including expert systems, self-organizing systems, machine learning systems, and artificial intelligence systems, may benefit from the use of neural networks, such as neural networks trained for pattern recognition. This is for purposes such as classifying one or more parameters, characteristics, or phenomena, assisting autonomous control, and other purposes. References to neural networks throughout this specification include feedforward neural networks, radial basis neural networks, self-organizing neural networks (e.g., Kohonen self-organizing neural networks), recurrent neural networks, modular neural networks, artificial neural networks, physical neural networks, multilayer neural networks, convolutional neural networks, hybrids of neural networks and other expert systems (e.g., hybrid fuzzy logic-neural network systems), autoencoder neural networks, stochastic neural networks, time-delay neural networks, convolutional neural networks, control feedback neural networks, radial basis neural networks, recurrent neural networks, Hopfield neural networks, Boltzmann machine neural networks. Recurrent neural networks, self-organizing map (SOM) neural networks, learning vector quantization (LVQ) neural networks, fully recurrent neural networks, simple recurrent neural networks, echo state neural networks, long short-term memory neural networks, bidirectional neural networks, hierarchical neural networks, probabilistic neural networks, genetic scale RNN neural networks, machine-based neural networks, associative neural networks, physical neural networks, instantaneous training neural networks, spike neural networks, neocognitron neural networks, dynamic neural networks, cascaded neural networks, neurofuzzy neural networks, constructive pattern generation neural networks, memory neural networks, hierarchical time-series memory neural networks, deep feedforward neural networks,Gated recurrent unit (GCU) neural networks, autoencoder neural networks, variational autoencoder neural networks, denoising autoencoder neural networks, sparse autoencoder neural networks, Markov chain neural networks, restricted Boltzmann machine neural networks, deep belief neural networks, deep convolutional neural networks, deconvolutional neural networks, deep convolutional inverse graphics neural networks, generative adversarial neural networks, liquid state machine neural networks, extreme learning machine neural networks, echo state neural networks, deep residual neural networks, support vector machine neural networks, neural Turing machine neural networks, and / or holographic associative memory neural networks, or hybrids or combinations of the above, or combinations with other expert systems such as rule-based systems, model-based systems (including those based on physical models, statistical models, flow-based models, biological models, biomimetic models, etc.).

[0075] In the embodiment, the exemplary neural network has cells assigned functions and requirements. In the embodiment, various examples of neural networks may include backfeed data / sensor cells, data / sensor cells, noise input cells, and hidden cells. Neural network components may also include stochastic hidden cells, spiked hidden cells, output cells, input / output matching cells, recursive cells, memory cells, different memory cells, kernels, and convolutional or pooled cells.

[0076] In this embodiment, an exemplary perceptron neural network can be connected to, integrated with, or interfaced with platform 100. The platform may further be associated with further neural network systems such as feedforward neural networks, radial-based neural networks, deep feedforward neural networks, recurrent neural networks, long / short-term memory neural networks, and gated recurrent neural networks. The platform may further be associated with further neural network systems such as autocoding neural networks, variational neural networks, denoising neural networks, sparse neural networks, Markov chain neural networks, and Hopfield network neural networks. The platform may also be associated with additional neural network systems such as Boltzmann machine neural networks, restricted Boltzmann machine neural networks, deep belief neural networks, deep convolutional neural networks, deconvolutional neural networks, and deep convolutional inverse graphics neural networks. This platform may also be associated with further neural network systems such as generative adversarial neural networks (GANs), liquid state machine neural networks, extreme learning machine neural networks, echo state neural networks (ESNs), deep residual neural networks, Kohonen neural networks, support vector machine neural networks, and neural Turing machine neural networks.

[0077] The aforementioned neural networks have various nodes or neurons, which perform various functions on inputs (including sensors, other data sources, and other nodes). These functions include weights, features, and feature vectors. Neurons include perceptrons and neurons that mimic biological functions (such as human touch, vision, taste, hearing, and smell). Continuous neurons, such as those with sigmoid activation functions, may be used in the context of various forms of neural networks, including those involving backpropagation.

[0078] In many embodiments, expert systems or neural networks are trained, for example, by a human operator or supervisor, or based on datasets, models, etc. Training involves presenting the neural network with one or more training datasets representing values ​​such as sensor data, event data, parameter data, and other data (including many types described herein), and one or more metrics indicating results such as process results, computation results, event results, activity results, etc. Training may include optimization training, which involves training the neural network to optimize one or more systems based on one or more optimization approaches, for example, Bayesian methods, parametric Bayesian classifier methods, k-nearest neighbor classifier methods, iterative methods, interpolation methods, Pareto optimization methods, and algorithmic methods. Feedback may be provided in the process of mutation and selection, for example, by a genetic algorithm that evolves one or more solutions based on feedback over a series of rounds.

[0079] In embodiments, multiple neural networks are deployed to a cloud platform that receives data streams and other inputs collected in one or more transactional environments (e.g., by mobile data collectors) and transmits them to the cloud platform via one or more networks. This includes the use of network coding to provide efficient transmission. The cloud platform can leverage large-scale parallel computing capabilities as needed and use multiple different types of neural networks (including modular, structurally adaptive, and hybrid) to perform prediction, classification, and control functions, and provide other outputs as described in relation to the expert systems disclosed throughout this specification. The different neural networks are configured to compete with each other (which may include using evolutionary algorithms, genetic algorithms, etc.) so that the appropriate type of neural network with the appropriate input set, weights, node types, and functions, etc., is selected, for example, by the expert system, for a particular task (related to a particular context, workflow, environmental process, system, etc.).

[0080] Embodiments, methods, and systems including expert systems or self-organizing functions described herein may utilize a feedforward neural network in which information moves in one direction. For example, information moves from a data input, such as a data source (any of the data sources referred to herein) relating to parameters related to at least one resource or transactional environment, through a series of neurons or nodes to an output. The data moves from the input node to the output node, without loops, and passing through one or more hidden nodes as needed. In embodiments, the feedforward neural network may consist of various types of units, such as binary MacCarloc-Pitts neurons (the simplest being a perceptron).

[0081] In embodiments, methods and systems including expert systems or self-organizing functions described herein may utilize encapsulated neural networks, for example, for predictive, classification, or control functions with respect to transactional environments related to one or more machine and automation systems described throughout this specification.

[0082] In embodiments, the expert systems or methods and systems with self-organizing capabilities described herein may utilize radial basis function (RBF) neural networks in situations involving interpolation in multidimensional space (e.g., optimization of data marketplaces, power generation systems, factory systems, and other multidimensional situations as described herein). In embodiments, each neuron in the RBF neural network stores an example from the training set as a "prototype." The linearity inherent in the functionality of this neural network offers the advantage that RBFs are not typically plagued by local minimum and maximum problems.

[0083] In embodiments, the methods and systems described herein, which involve expert systems or self-organizing capabilities, may utilize radial basis function (RBF) neural networks that employ a distance criterion relative to a center (e.g., a Gaussian function). Radial basis functions can be applied in multilayer perceptrons as alternatives to hidden layers, such as sigmoid hidden layer transfer functions. An RBF network may have a two-layer structure, for example, where the input is mapped to each RBF in the hidden layer. In embodiments, the output layer may consist of a linear combination of hidden layer values ​​representing, for example, the average predicted output. The values ​​in the output layer provide an output equivalent to or similar to a regression model in statistics. In classification problems, the output layer is a sigmoid function of the linear combination of hidden layer values, representing the posterior probability. In either case, performance is often improved by shrinkage techniques, such as ridge regression in classical statistics. This corresponds to prior belief in a small prior distribution of parameter values ​​(and therefore a smooth output function) in a Bayesian framework. Because the RBF network only modifies a linear mapping from the hidden layer to the output layer during the learning process, it can avoid local minima. Linearity guarantees that the error surface is a quadratic function and therefore has a single minimum value. In regression problems, this can be found in a single matrix operation. In classification problems, the fixed nonlinearity introduced by the sigmoid output function can be handled using methods such as iterative reweighted least squares. RBF networks can utilize kernel methods such as support vector machines (SVMs) and Gaussian processes (where RBF is the kernel function). By using a nonlinear kernel function, input data can be projected into a space where learning problems can be solved with linear models.

[0084] In embodiments, an RBF neural network may include an input layer, a hidden layer, and a sum layer. In the input layer, one neuron appears for each predictor variable. For categorical variables, N-1 neurons are used (where N is the number of categories). In embodiments, the input neurons can be standardized by subtracting the median and dividing by the interquartile range. The input neurons then supply values ​​to each neuron in the hidden layer. The hidden layer may use a variable number of neurons (determined by the learning process). Each neuron consists of a radial basis function (RBF) centered on a point with the same dimensions as the number of predictor variables. The spread (e.g., radius) of the RBF function may differ for each dimension. The center point and spread are determined by learning. When the input value vector from the input layer is presented, the hidden neurons calculate the Euclidean distance between the test cases and their own center point and apply an RBF kernel function (e.g., using the spread value) to this distance. The resulting values ​​are then passed to the sum layer. In the sum layer, the output values ​​from neurons in the hidden layer are multiplied by the weights associated with that neuron and added together with the weighted values ​​of other neurons. This sum is the output. In classification problems, one output is generated for each target category (using separate weights and sum units). The output value for a category is the probability that the case being evaluated belongs to that category. In learning an RBF, various parameters are determined, such as the number of neurons in the hidden layer, the center coordinates of each hidden layer function, the spread of each function in each dimension, and the weights transmitted from the output to the sum layer. Clustering algorithms (such as k-means clustering) and evolutionary algorithms are used for learning.

[0085] In some embodiments, a recurrent neural network may have time-varying real-valued activations (outputs) (simply values ​​of 0 or greater). Each connection has a mutable real-valued weight. Some nodes are called labeled nodes, some are called output nodes, and others are called hidden nodes. In supervised learning in a discrete-time setting, the training sequence of real-valued input vectors can be the activation sequence of input nodes (processing the input vectors one by one). At each time step, each non-input unit calculates its current activation value as a nonlinear function of the weighted sum of the activation values ​​of all the units it connects to. The system can explicitly activate some output units at a particular time step, independently of the input signal.

[0086] Embodiments, methods, and systems with expert systems or self-organizing capabilities described herein may utilize self-organizing neural networks, such as Kohonen self-organizing neural networks, for the visualization of data views, such as low-dimensional views of high-dimensional data. Self-organizing neural networks can apply competitive learning to input datasets from one or more sensors or other data inputs related to a transactional environment (including any machines or components related to the transactional environment). In embodiments, self-organizing neural networks may be used to identify structures within data (e.g., unlabeled data). For example, this applies to data sensed from various data sources or sensors within or around a transactional environment where the source of the data is unknown (e.g., an event may originate from one of several unknown sources). Self-organizing neural networks organize structures and patterns within the data so that they can be recognized, analyzed, and labeled. For example, they can identify market behavioral structures as corresponding to other events or signals.

[0087] Embodiments, methods, and systems described herein that include expert systems or self-organizing functions may utilize recurrent neural networks, which enable bidirectional data flow, for example, when connected units (such as neurons or nodes) form directed cycles. Such networks may be used to model or represent the dynamic temporal behavior observed in dynamic systems (e.g., the wide variety of automated systems, machines, and devices described throughout this specification). Examples include automated agents that interact with markets for purposes such as data collection, spot market trading testing, and trade execution. Dynamic system behavior involves complex interactions that users wish to understand, predict, control, and / or optimize. For example, recurrent neural networks can be used to predict market states involving dynamic processes and behaviors (such as changes in the state of resources being traded or resources enabling the market in the trading environment). In embodiments, a recurrent neural network may use its internal memory to process input sequences from other nodes and sensors, and from or relating to the trading environment, among other data inputs, as described herein. In embodiments, recurrent neural networks may also be used for pattern recognition, such as the recognition of machines, components, agents, or other items based on behavioral signatures, profiles, feature vectors (such as audio files or images), or similarities. As a non-limiting example, a recurrent neural network may recognize shifts in the operating modes of a market or machine by learning to classify shifts in operating modes from a training dataset consisting of data streams from data sources of one or more sensors applied to or relating to one or more resources.

[0088] Embodiments, methods, and systems involving expert systems or self-organizing capabilities described herein may utilize modular neural networks. A modular neural network may consist of a set of independent neural networks (e.g., the various types of neural networks described herein) coordinated by an intermediary. Each independent neural network within the modular neural network may operate on individual inputs and perform subtasks that constitute the task intended to be performed by the modular network as a whole. For example, a modular neural network may include a recurrent neural network for pattern recognition (such as recognizing the type of machine or system detected by one or more sensors provided as input channels to the modular network) and an RBF neural network for optimizing the operation of the understood machine or system. The intermediary receives and processes inputs from each individual neural network and generates outputs for the modular neural network (e.g., appropriate control parameters, state predictions).

[0089] Any pair, triplet, or more combinations of the various neural network types described herein are included in this disclosure. This includes combinations in which an expert system uses one neural network to recognize patterns (e.g., patterns indicating problems or failure conditions) and another neural network to self-organize activities or workflows based on the recognized patterns (e.g., to provide outputs that govern the autonomous control of the system in response to the recognized conditions or patterns). It also includes combinations in which an expert system uses one neural network to classify items (e.g., identification of machines, components, or operating modes) and another neural network to predict the state of the items (e.g., failure state, operating state, predicted state, maintenance state, etc.). Modular neural networks also include situations where an expert system uses one neural network to determine a state or context (such as a machine, process, workflow, market, storage system, network, or data acquisition device) and another neural network to self-organize processes related to that state or context (e.g., data storage processes, network coding processes, network selection processes, data marketplace processes, power generation processes, manufacturing processes, refining processes, drilling processes, boring processes, or other processes described herein).

[0090] Methods and systems including expert systems or self-organizing functions described herein may use physical neural networks that perform or simulate neural behavior using one or more hardware elements. In embodiments, one or more hardware neurons may be configured to stream voltage values, current values, etc., representing sensor data, such as calculating information from analog sensor inputs representing energy consumption, energy production, etc., when one or more machines supply or consume energy for one or more transactions. One or more hardware nodes may be configured to stream output data resulting from the activity of the neural network. Hardware nodes, including one or more chips, microprocessors, integrated circuits, programmable logic controllers, application-specific integrated circuits, field-programmable gate arrays, etc., may be provided to optimize machines that produce or consume energy, or to optimize other parameters in any of the types of neural networks described herein. Hardware nodes may include hardware for accelerating computations (such as dedicated processors that perform basic or more advanced computations on input data and provide outputs, dedicated processors that filter or compress data, dedicated processors that decode data, and dedicated processors that compress specific files or data types (e.g., image data, video streams, acoustic signals, thermal images, heatmaps, etc.)). Physical neural networks can be embodied in data collectors. These include those that are reconfigurable by switching and routing inputs in various configurations (for example, providing different neural network configurations within the data collector to handle different input types) (the switching and configuration are arbitrarily performed under the control of an expert system, including a software-based neural network located on or remotely on the data collector).A physical, or at least partially physical, neural network may include physical hardware nodes located within a machine, data storage system, distributed ledger, mobile device, server, cloud resource, or storage system for storing data in a transactional environment. This is to accelerate input / output functions to or from one or more storage elements that supply or receive data to or from the neural network. A physical, or at least partially physical, neural network may also include physical hardware nodes located within a network, such as in data transmission within, to, or from an industrial environment. For example, to accelerate input / output functions to or from one or more network nodes within the network, or to accelerate relay functions. In embodiments of a physical neural network, the function of a neural synapse can be emulated using electrically adjustable resistive materials. In embodiments, the physical hardware emulates neurons, and the software emulates the neural network between neurons. In embodiments, the neural network complements conventional algorithmic computers. They are highly versatile and can be trained to perform appropriate functions without requiring any instructions, such as classification, optimization, pattern recognition, control, selection, and evolution.

[0091] In embodiments, the methods and systems described herein, which involve expert systems or self-organizing capabilities, may utilize multilayer feedforward neural networks for the classification of complex patterns such as one or more items, phenomena, modes, or states. In embodiments, the multilayer feedforward neural network may be trained by optimization techniques, such as genetic algorithms, which explore a large and complex choice space to find a global solution that is optimal or near-optimal. For example, one or more genetic algorithms may be used to train a multilayer feedforward neural network to classify complex phenomena. This includes the recognition of complex operating modes of machines (such as modes involving complex interactions between machines (including interference effects, resonance effects, etc.), modes involving nonlinear phenomena, and modes involving critical failures where multiple simultaneous failures occur, making root cause analysis difficult). In embodiments, the multilayer feedforward neural network may be used to classify the results of market surveillance, including surveillance systems such as automated agents operating within a market, as well as surveillance resources that enable the market, such as computing, networking, energy, data storage, energy storage, and other resources.

[0092] In embodiments, the methods and systems described herein, which involve expert systems or self-organizing capabilities, may utilize feedforward-backpropagation multilayer perceptron (MLP) neural networks. For example, they are used to process one or more remote sensing applications, such as taking input from sensors distributed across various trading environments. In embodiments, the MLP neural network is used to classify trading and resource environments (e.g., spot markets, futures markets, energy markets, renewable energy credit (REC) markets, network markets, advertising markets, spectrum markets, ticket markets, reward markets, computing markets, and other markets mentioned throughout this specification), as well as the physical resources and environments that generate them (including classification of energy resources (including renewable energy environments, mining environments, exploration environments, drilling environments, etc.), geological structures (including subsurface and surface structures), materials (including fluids, minerals, metals, etc.), and other issues). This may include fuzzy classification.

[0093] Embodiments, methods, and systems described herein that involve expert systems or self-organizing capabilities may utilize structure-adaptive neural networks, where the structure of the neural network is adapted based on rules, detected states, contextual parameters, etc. For example, if a neural network, after some learning, does not converge to a solution such as item classification or prediction when operating on a series of inputs, the neural network may be modified, for example, by changing from a feedforward neural network to a recurrent neural network, or by switching the data path between subsets of nodes from unidirectional to bidirectional. Structural adaptation may occur under the control of an expert system, such as by triggering adaptation in the event of a trigger, rule, or event (e.g., recognition of a threshold occurrence (e.g., failure to converge to a solution within a given time), or recognition of a phenomenon requiring a different or additional structure (e.g., recognition that the system is changing dynamically or non-linearly)). For example, when an expert system receives a directive that a continuously variable transmission is used to drive generators, turbines, etc., in the system being analyzed, it can switch from a simple structure like a feedforward neural network to a more complex structure like a recurrent neural network or a convolutional neural network.

[0094] Embodiments, methods, and systems with expert systems or self-organizing capabilities described herein may utilize autoencoders, autoassociators, or diabolo neural networks, which may be similar to multilayer perceptron (MLP) neural networks, such as having input layers, output layers, and one or more hidden layers connecting them. However, the output layer of an autoencoder has the same number of units as the input layer, and the purpose of an MLP neural network is to reconstruct its own input (rather than simply outputting a target value). Thus, autoencoders can operate as unsupervised learning models. For example, autoencoders are used for unsupervised learning of efficient coding for dimensionality reduction, learning of data generation models, etc. In embodiments, autoencoder neural networks may be used to self-learn efficient network coding for transmitting analog sensor data from a machine through one or more networks, or for transmitting digital data from one or more data sources. In embodiments, autoencoder neural networks may be used to self-learn efficient storage techniques for storing data streams.

[0095] Embodiments, methods, and systems with expert systems or self-organizing capabilities described herein may utilize stochastic neural networks (PNNs). In embodiments, a PNN may include a multi-layer (e.g., four-layer) feedforward neural network, where the layers include an input layer, a hidden layer, a pattern / sum layer, and an output layer. In one embodiment of the PNN algorithm, the parent probability distribution function (PDF) for each class is approximated by a Parzen window and / or a nonparametric function. The class probability of a new input can then be estimated using the PDF for each class, and Bayes' theorem can be applied to assign it to the class with the highest posterior probability (e.g.). The PNN embodies a Bayesian network and may utilize statistical algorithms or analytical techniques such as kernel Fisher discriminant analysis. PNNs may be used for classification and pattern recognition in any of the extensive embodiments disclosed herein. As a non-limiting example, a stochastic neural network may be used to predict engine failure conditions based on the collection of data inputs from engine sensors and instruments.

[0096] Embodiments, methods, and systems with expert systems or self-organizing capabilities described herein may use a time-delayed neural network (TDNN) that has a forward transfer architecture for time-series data and recognizes ordinal position-independent features. In embodiments, to account for time shifts in the data, delays are added to one or more inputs or between one or more nodes so that multiple data points (from different points in time) are analyzed together. The time-delayed neural network may form part of a larger pattern recognition system, such as using a perceptron network. In embodiments, the TDNN may be trained by supervised learning, such as training connection weights under backpropagation or feedback. In embodiments, the TDNN is used to process sensor data from different streams, such as velocity data streams, acceleration data streams, temperature data streams, and pressure data streams, and time delays are used to temporally align the data streams to aid in pattern understanding, which involves understanding various streams (e.g., changes in price patterns in a spot market or futures market).

[0097] In embodiments, the methods and systems described herein, which involve expert systems or self-organizing capabilities, may utilize convolutional neural networks (sometimes referred to as CNNs, ConvNets, shift-invariant neural networks, or spatially invariant neural networks), whose units are connected in a pattern similar to the visual cortex of the human brain. Neurons respond to stimuli in limited spatial regions called receptive fields. Receptive fields may partially overlap, and collectively they cover an entire (e.g., visual) cortex. Node responses are mathematically computed, such as by convolutional operations using multilayer perceptrons with minimal preprocessing. Convolutional neural networks are used for recognition in image and video streams, for example, by camera systems installed on mobile data acquisition devices such as drones or mobile robots to recognize the type of machine in a vast environment. In embodiments, convolutional neural networks are used to provide recommendations based on data inputs, including sensor inputs and other contextual information, for example, by recommending routes for mobile data acquisition devices. In the examples, convolutional neural networks may be used for input processing, such as natural language processing of instructions provided by one or more parties involved in a workflow within an environment. In the examples, the convolutional neural network is deployed using a large number of neurons (e.g., 100,000, 500,000 or more), multiple layers (e.g., 4, 5, 6 or more), and a large number of parameters (e.g., millions). One or more convolutional neural networks may be used.

[0098] In the embodiments, the methods and systems described herein, with expert systems or self-organizing capabilities, can utilize regulatory feedback networks and can be used, for example, to recognize emergent phenomena (such as new types of behavior previously not understood in transactional environments).

[0099] In embodiments, the methods and systems described herein, including expert systems or self-organizing functions, may utilize self-organizing maps (SOMs) with unsupervised learning. A group of neurons learns to map points in the input space to coordinates in the output space. The input space has different dimensions and topology from the output space, and the SOM preserves these when mapping phenomena into groups.

[0100] Embodiments, methods, and systems including expert systems or self-organizing functions described herein may utilize learned vector quantized neural networks (LVQs). Representative examples of classes are parameterized in distance-based classification schemes, along with appropriate distance measurement methods.

[0101] In embodiments, the methods and systems described herein, including expert systems or self-organizing functions, may utilize an echo state network (ESN) consisting of a recurrent neural network with loosely connected random hidden layers. The weights of the output neurons are modified (e.g., the weights are learned based on feedback). In embodiments, the ESN may be used to process time-series patterns, for example, to recognize patterns of market-related events (such as patterns of price fluctuations in response to stimuli).

[0102] In embodiments, the methods and systems described herein, which involve expert systems or self-organizing capabilities, may utilize bidirectional recurrent neural networks (BRNNs). For example, a finite-value sequence (such as voltage values ​​from a sensor) is used to predict or label each element of the sequence based on both the past and future context of the elements. This can be achieved, for example, by adding the outputs of two RNNs: one that processes from left to right and another that processes from right to left. The combined output is a predicted value of a target signal, such as one provided by a teacher or supervisor. Bidirectional RNNs can be combined with long-term short-term memory (LSTM) RNNs.

[0103] Embodiments, methods, and systems described herein that involve expert systems or self-organizing capabilities may utilize a hierarchical RNN that connects elements in various ways to decompose hierarchical behaviors (such as useful subprograms). In embodiments, the hierarchical RNN may be used to manage one or more hierarchical templates for data collection in a transactional environment.

[0104] Methods and systems including expert systems or self-organizing functions described herein may utilize stochastic neural networks, which may introduce random fluctuations into the network. Such random fluctuations can be considered a form of statistical sampling, such as Monte Carlo sampling.

[0105] In the embodiments, the methods and systems described herein, including expert systems or self-organizing functions, may utilize genetically scaled recurrent neural networks. In such embodiments, a sequence is decomposed into multiple scales, and each scale uses an RNN (often an LSTM) that determines the major length between any two consecutive points. The first-order scale consists of a regular RNN, the second-order scale consists of all points separated by two indices, and so on. The Nth-order RNN connects the first and last nodes. The outputs from the various scales are treated as members of a committee, and the associated scores are used genetically for subsequent iterations.

[0106] Embodiments, methods, and systems involving expert systems or self-organizing capabilities described herein may utilize a Committee of Machines (CoM). A CoM is a collection of different neural networks that collectively "vote" on specific examples. Because neural networks can fall into local optima, even starting with the same architecture and training, setting initial weights to randomly different values ​​often yields different results. CoMs tend to stabilize the results.

[0107] Embodiments, methods, and systems with expert systems or self-organizing capabilities described herein may utilize associative neural networks (ASNNs), such as extensions of committees of machines combining multiple feedforward neural networks and k-nearest neighbors (kNNs). In kNNs, correlations between ensemble responses may be used as a measure of distance between cases under analysis. This corrects for bias in the neural network ensemble. Associative neural networks may have memories that may match the training set. When new data becomes available, the network immediately improves its predictive ability without retraining, providing data approximation (self-learning). Another important feature of ASNNs is the possibility of interpreting neural network results by correlation analysis between data instances in the model space.

[0108] Embodiments, methods, and systems described herein that involve expert systems or self-organizing capabilities may use an immediately-trained neural network (ITNN) that maps the weights of hidden and output layers directly from training vector data.

[0109] Embodiments, methods, and systems including expert systems or self-organizing functions described herein may use spike neural networks that can explicitly consider the timing of inputs. Network inputs and outputs are represented as a series of spikes (such as delta functions or more complex shapes). SNNs process information in the time domain (e.g., time-varying signals such as signals related to the dynamic behavior of markets or trading environments). These are often implemented as recursive networks.

[0110] In the embodiments, the methods and systems described herein, including specialized systems or self-organizing functions, may address nonlinear multivariate behavior and utilize dynamic neural networks that include learning of time-dependent behaviors such as transients and delay effects. Transients may include fluctuations in market variables such as price, available quantity, and available trading partners.

[0111] In embodiments, cascaded correlation is used as both an architecture and a supervised learning algorithm to complement weight tuning in fixed-topology networks. Cascaded correlation starts with a minimal network, automatically learns, and builds a multi-layered structure by adding new hidden units one by one. As a new hidden unit is added to the network, its input weights are fixed. This unit becomes a permanent feature detector within the network and can be used for output generation or building more complex feature detectors. The cascaded correlation architecture learns quickly, determines its own size and topology, and maintains the structure it has built. This is maintained even as the training set changes and does not require backpropagation.

[0112] Embodiments, methods, and systems with expert systems or self-organizing capabilities described herein may utilize neurofuzzy networks, such as by incorporating a fuzzy inference system into the artificial neural network itself. Depending on the type, multiple layers may simulate processes related to fuzzy inference (such as fuzzification, inference, aggregation, and defuzzification). The advantage of incorporating a fuzzy system into the general structure of a neural network is that the parameters of the fuzzy system can be determined using available learning techniques.

[0113] Embodiments, methods, and systems described herein that involve expert systems or self-organizing capabilities may utilize constructive pattern-generating networks (CPPNs), which are variants of associative neural networks (ANNs) with different sets of activation functions and methods of application. While typical ANNs often contain only sigmoid functions (and sometimes Gaussian functions), CPPNs may contain both function types and many other functions. Furthermore, CPPNs are applicable to the entire space of inputs and achieve complete image representations. As a composite of functions, CPPNs can encode images at virtually infinite resolution and sample them at optimal resolutions for specific displays.

[0114] This type of network can add new patterns without retraining. Embodiments, methods, and systems of expert systems or self-organizing capabilities described herein may utilize one-shot associative memory networks by constructing specific memory structures (e.g., assigning each new pattern to an orthogonal plane using adjacently connected hierarchical arrays).

[0115] Embodiments, methods, and systems including expert systems or self-organizing functions described herein may utilize hierarchical spatiotemporal memory (HTM) neural networks that incorporate structural and algorithmic properties of the neocortex. HTM may employ biomimetic models based on memory prediction theory. HTM may be used to discover and infer higher-order causes of observed input patterns and sequences.

[0116] (Machine learning system) In embodiments, the machine learning system can train models such as predictive models (e.g., various types of neural networks, regression-based models, and other machine learning models). In embodiments, training can be supervised, semi-supervised, or unsupervised. In embodiments, training can be performed using training data collected or generated for training purposes.

[0117] A facility output model (or predictive model) may be a model that takes facility attributes and outputs one or more predictions about the facility's output or other outputs. Examples of predictions include the amount of energy the facility generates, the amount the facility processes, the amount of data the network can transfer, the amount of data that can be stored, the price of parts and services (including those supplied to or provided by the facility), the profit from completing a given task, and the cost of performing an action. In any case, the machine learning system arbitrarily trains the model based on training data. In embodiments, the machine learning system may receive vectors containing facility attributes (e.g., facility type, facility capacity, achievement goals, constraints or rules applicable to the use of resources or facilities), person attributes (e.g., roles, components managed), and results (e.g., energy generated, computing tasks completed, financial results, etc.). Each vector corresponds to a respective result and the attributes of the respective facility and the respective action that produced that result. The machine learning system receives these vectors and generates a predictive model based on them. In embodiments, the machine learning system may store the predictive model in a model data store.

[0118] In some embodiments, the system can also perform learning based on feedback received (also known as reinforcement learning). In some embodiments, the machine learning system receives a set of circumstances that led to a prediction (e.g., facility attributes, model attributes, etc.) and results related to the facility, and updates the model based on the feedback.

[0119] In some embodiments, training may be provided from a training dataset created by observing the behavior of multiple individuals, such as facility managers who manage facilities with diverse capabilities and are involved in various contexts and situations. This may include the use of process automation by robots to learn a training dataset of human-computer interaction with interfaces, such as graphical user interfaces of one or more computer programs, including dashboards, control systems, and other systems used to manage energy and computing management facilities.

[0120] (Artificial Intelligence (AI) System) In this embodiment, the AI ​​system utilizes a predictive model to make predictions about a facility. Examples of predictions include predictions related to facility inputs (such as available energy, energy costs, computing resource costs, and network capacity, as well as various market information such as pricing information for the end-user market), predictions related to facility components or systems (such as performance predictions, maintenance predictions, uptime / downtime predictions, and capacity predictions), predictions related to facility functions or workflows (including conditions and states that result in following one of several different possible paths in a workflow, process, etc.), predictions related to facility outputs, and others. In this embodiment, the AI ​​system receives a facility identifier. In response to the facility identifier, the AI ​​system may retrieve attributes corresponding to that facility. In some embodiments, the AI ​​system may retrieve facility attributes from a graph. Furthermore, or alternatively, the AI ​​system may retrieve facility attributes from a facility record corresponding to a facility identifier and person attributes from a person record corresponding to a person identifier.

[0121] Examples of additional attributes that can be used to make predictions about a facility or associated system processes include relevant facility information, owner objectives (including financial objectives), client objectives, and many other additional or alternative attributes. In an embodiment, the AI ​​system may output a score for each predictability. Each prediction corresponds to a possible outcome. For example, if a prediction model is used to determine the possibility that a facility's hydroelectric power source will produce 5 MW of power, the prediction model may output scores for the "produces" and "does not produce" outcomes. The AI ​​system then selects the outcome with the highest score as the prediction. Alternatively, the AI ​​system may output each score to the requesting system.

[0122] (Intelligence Service System) Figure 3 shows an example of an intelligence system 300 (also referred to as the “intelligence service,” the “intelligence service system,” or the “intelligence system”) according to one embodiment of the present disclosure. In the embodiment, the intelligence system 300 provides a framework for providing intelligence services to one or more intelligence service clients 336. In some embodiments, the framework of the intelligence system 300 is adaptable to be at least partially replicated in each intelligence client 336 (e.g., an enterprise access tier, a wallet system, a market orchestration system, a digital lending system, an asset-backed tokenization system, and / or similar). In these embodiments, each client 336 may include some or all of the functionality of the intelligence system 300, thereby allowing the intelligence system 300 to adapt to specific functionality performed by the subsystems of the intelligence clients. Furthermore, or alternatively, in some embodiments, the intelligence system 300 may be implemented as a set of microservices. This allows different intelligence clients 336 to leverage the intelligence system 300 through one or more APIs exposed to the intelligence clients. In these embodiments, the intelligence system 300 may be configured to perform various types of intelligence services that are adaptable to different intelligence clients 336. In any configuration, the intelligence service client 336 provides intelligence requests to the intelligence system 300, which request the execution of specific intelligence tasks (e.g., decisions, recommendations, reports, instructions, classification, predictions, training actions, NLP requests, etc.). In response, the intelligence system 300 performs the requested intelligence task and returns a response to the intelligence service client 336.Furthermore, or alternatively, in some embodiments, the intelligence system 300 may be implemented using one or more dedicated chips configured to provide AI-assisted microservices such as image processing, diagnostics, position and orientation, chemical analysis, and data processing. Examples of AI-enabled chips are discussed elsewhere in this specification.

[0123] In an embodiment, the intelligence system 300 may include an intelligence service controller 302 and an artificial intelligence (AI) module 304. In an embodiment, the AI ​​system 300 receives intelligence requests and data necessary to process the requests from an intelligence service client 336. In response to the requests and specific data, one or more relevant AI modules 304 perform intelligence tasks and output an "intelligence response". Examples of responses from the intelligence module 304 include decisions (e.g., control commands, suggested actions, machine-generated text, and / or similar), predictions (e.g., predicted meaning of a text fragment, predicted results related to a suggested action, predicted failure state, etc.), classifications (e.g., classification of objects in an image, classification of speech content, classified failure state based on sensor data, etc.), and / or other appropriate outputs from the AI ​​system.

[0124] (Artificial Intelligence Module) In this embodiment, the artificial intelligence module 304 may include a machine learning module 312, a rule-based module 328, an analysis module 318, an RPA module 316, a digital twin module 320, a machine vision module 322, a natural language processing module 324, and / or a neural network module 314. Note that the aforementioned artificial intelligence modules are non-limiting examples, and some modules may be encompassed by or utilized by other artificial intelligence modules. For example, the NLP module 324 and the machine vision module 322 may utilize different neural networks, which are part of the neural network module 314, in performing their respective functions.

[0125] Furthermore, it should be noted that in certain scenarios, the artificial intelligence module 304 itself may become the intelligence client 336. For example, the rule-based intelligence module 328 may request an intelligence task from the machine learning module 312 or the neural network module 314, such as classifying objects appearing in a video and / or requesting their movement. In this example, the rule-based intelligence module 328 becomes the intelligence service client 336, which decides whether to perform a specific action using the classification results. In another example, the machine vision module 322 may request a digital twin of a particular environment from the digital twin module 320, and the ML module 312 may request specific data from the digital twin as features to train a machine learning model trained for that particular environment.

[0126] In embodiments, intelligence tasks may require specific types of data to respond to requests. For example, a machine vision task may require one or more images (and possibly other data) to classify objects appearing in an image or a series of images and to identify features within the series of images (such as item locations, the presence of faces, symbols, or instructions, facial expressions, behavioral parameters, or state changes). Another example is an NLP task which requires audio data and / or text data (and possibly other data) to determine the meaning and other elements of the audio and / or text. Yet another example is an AI-based control task (e.g., decisions about robot movement) which may require environmental data (e.g., maps, coordinates of known obstacles, images, and / or similar) and / or a motion plan to determine how to control the robot's movements. In a platform-level example, an analytics-based reporting task may require data from multiple different databases to generate a report. Thus, in embodiments, tasks that the intelligence system 300 can perform may require, or benefit from, specific intelligence service inputs 332. In some embodiments, the intelligence system 300 may be configured to receive and / or request specific data from an intelligence service input 332 in order to perform a corresponding intelligence task. Alternatively, a requesting intelligence service client 336 may provide specific data within its request. For example, the intelligence system 300 may expose one or more APIs to the intelligence client 336, thereby allowing the requesting client 336 to provide specific data within its request via the APIs. Examples of intelligence service inputs include, but are not limited to, sensors providing sensor data, video streams, audio streams, databases, data feeds, human input, and / or other appropriate data.

[0127] In an embodiment, the intelligence module 304 includes and provides access to an ML module 312 that can be integrated into or accessed by one or more intelligence clients 336. In an embodiment, the ML module 312 can provide machine learning-based capabilities, features, functions, and algorithms for use by the intelligence service client 336, such as training ML models, leveraging ML models, enhancing ML models, performing various clustering techniques, feature extraction, and / or similar actions. As an example, the machine learning module 312 can provide machine learning computing, data storage, and feedback infrastructure to a simulation system (e.g., as described above). The machine learning module 312 can also work in conjunction with the rule-based module 328, the machine vision module 322, the RPA module 316, and / or other similar modules.

[0128] The machine learning module 312 can define one or more machine learning models for performing analysis, simulation, decision-making, and predictive analysis related to data processing, data analysis, simulation creation, and simulation analysis of one or more components or subsystems of the intelligence service client 336. In embodiments, a machine learning model is an algorithm and / or statistical model that performs a particular task without using explicit instructions, and instead relies on patterns and inference. A machine learning model builds one or more mathematical models on training data to make predictions and / or decisions without being explicitly programmed to perform a particular task. In exemplary implementations, a machine learning model may perform classification, prediction, regression, clustering, anomaly detection, recommendation generation, and / or other tasks.

[0129] In embodiments, a machine learning model can perform various types of classification based on input data. Classification is a predictive modeling problem in which class labels are predicted for given examples of input data. For example, a machine learning model can perform binary classification, multi-class classification, or multi-label classification. In embodiments, a machine learning model may output a "confidence score" indicating the confidence level of each classification of the input to a given class. In embodiments, the confidence scores can be compared against one or more thresholds to perform discrete category prediction. In embodiments, only a specific number of classes (e.g., one) with relatively high confidence scores can be selected, and discrete category prediction can be performed.

[0130] In embodiments, a machine learning model can output probabilistic classifications. For example, a machine learning model can predict a probability distribution on a set of classes for a sample input. Thus, instead of outputting only the most likely class to which the sample input should belong, the machine learning model can output the probability that the sample input belongs to each class. In embodiments, the probability distribution across all possible classes sums to 1. In embodiments, a Softmax function, or other types of functions or layers, can be used to transform the set of real values ​​associated with each possible class into a set of real values ​​in the range (0,1) that sums to 1. In embodiments, discrete category prediction can be performed by comparing the probabilities provided by the probability distribution with one or more thresholds. In embodiments, discrete category prediction can be performed by selecting only a specific number of classes (e.g., one) with the relatively highest predicted probability.

[0131] In the examples, the machine learning model can perform regressions and provide output data in a continuous numerical format. For example, the machine learning model can perform linear regression, polynomial regression, or nonlinear regression. As mentioned above, in the examples, the Softmax function or other function or layer can be used to compress a set of real values ​​associated with two or more possible classes into a set of real values ​​in the range (0,1) that sum to 1. For example, the machine learning model can perform linear regression, polynomial regression, or nonlinear regression. Specifically, the machine learning model can perform simple regression or multiple regression. As mentioned above, in some implementations, the Softmax function or other function or layer can be used to compress a set of real values ​​associated with two or more possible classes into a set of real values ​​in the range (0,1) that sum to 1.

[0132] In some embodiments, a machine learning model can perform various types of clustering. For example, a machine learning model may identify one or more predefined clusters that the input data is most likely to correspond to. In some embodiments where a machine learning model performs clustering, the machine learning model may be trained using unsupervised learning techniques.

[0133] In some embodiments, machine learning models can perform anomaly detection or outlier detection. For example, a machine learning model may identify input data that does not fit expected patterns or other characteristics (e.g., those previously observed from past input data). Anomaly detection can be used, for example, to detect fraud or system failures.

[0134] In some embodiments, a machine learning model can provide output data in the form of one or more recommendations. For example, a machine learning model can be incorporated into a recommendation system or engine. As an example, given input data describing past results for a particular entity (such as scores indicating success or satisfaction, rankings, or ratings), a machine learning model can output suggestions or recommendations for additional entities that are expected to produce desirable results based on the past results.

[0135] As mentioned above, a machine learning model can be one or more of various types of machine learning models, or it can include several of them. Examples of such different types of machine learning models are shown below for illustrative purposes. One or more of the example models described below (for example, in combination) can be used to provide output data in response to input data. Additional models other than those shown below are also available.

[0136] In some embodiments, the machine learning model may be one or more classification models, such as a linear classification model or a quadratic classification model, or may include such models. The machine learning model may be one or more regression models, such as a simple linear regression model, a multilinear regression model, a logistic regression model, a stepwise regression model, a multivariate adaptive regression spline, or a locally estimated scatter plot smoothing model, or may include such models.

[0137] In some examples, a machine learning model may be or contain one or more decision tree-based models, such as classification and / or regression trees, chi-squared auto-interaction detection decision trees, decision stamps, or conditional decision trees.

[0138] A machine learning model may include one or more kernel machines. In some embodiments, a machine learning model may include one or more support vector machines. A machine learning model may be or include one or more instance-based learning models, such as a learning vector quantization model, a self-organizing map model, or a locally weighted learning model. In some embodiments, a machine learning model may be or include one or more nearest neighbor models, such as a k-nearest neighbor classification model or a k-nearest neighbor regression model. A machine learning model may be or include one or more Bayesian models, such as a naive Bayes model, a Gaussian naive Bayes model, a polynomial naive Bayes model, an averaged one-dependency estimator, a Bayesian network, a Bayesian belief network, or a hidden Markov model.

[0139] A machine learning model can include one or more clustering models, such as k-means clustering models, k-medians clustering models, expectation maximization models, and hierarchical clustering models.

[0140] In some embodiments, the machine learning model may perform one or more dimensionality reduction techniques, such as principal component analysis, kernel principal component analysis, graph-based kernel principal component analysis, principal component regression analysis, partial least squares regression analysis, summon mapping, multidimensional scaling, projection search, linear discriminant analysis, mixed discriminant analysis, quadratic discriminant analysis, generalized discriminant analysis, flexible discriminant analysis, and autocoding.

[0141] In some embodiments, the machine learning model may implement or apply one or more reinforcement learning techniques, such as Markov decision processes, dynamic programming, Q-function or Q-learning, value function approaches, deep Q-networks, differentiable neural computers, asynchronous advantage actor-critical, and deterministic policy gradient methods.

[0142] In an embodiment, the artificial intelligence module 304 may include and / or provide access to the neural network module 314. In an embodiment, the neural network module 314 is configured to train, deploy, and / or utilize an artificial neural network (or “neural network”) on behalf of the intelligence service client 336. In this specification, the term machine learning model may include a neural network, and therefore, the neural network module 314 may be part of the machine learning module 312. In an embodiment, the neural network module 314 may be configured to train a neural network that may be used by the intelligence client 336. Non-limiting examples of different types of neural networks may include any neural network type described throughout this disclosure and the documents incorporated by reference. This includes convolutional neural networks (CNNs), deep convolutional neural networks (DCNs), feedforward neural networks (including deep feedforward neural networks), recurrent neural networks (RNNs) (including but not limited to gated RNNs), and long short-term memory (LSTM) neural networks. It also includes hybrids or combinations of these, which unfold in serial, parallel, acyclic (e.g., directed graph-based) flows, and / or more complex flows including intermediate decision nodes, recursive loops, etc. Here, a particular type of neural network receives input from a data source or another neural network and provides an output that is included in the input set of another neural network until the flow is complete and the final output is provided. In embodiments, the neural network module 314 may be utilized by other artificial intelligence modules 304, such as the machine vision module 322, the NLP module 324, the rule-based module 328, and the digital twin module 320. Examples of applications of the neural network module 314 are described throughout this disclosure.

[0143] A neural network consists of a group of connected nodes, also known as neurons or perceptrons. A neural network can consist of one or more layers. A neural network with multiple layers is called a "deep" network. A deep network can include an input layer, an output layer, and one or more hidden layers placed between the input and output layers. The nodes in a neural network are either connected or not fully connected.

[0144] In some embodiments, the neural network may be or include one or more feedforward neural networks. In a feedforward network, connections between nodes do not form cycles. For example, each connection may connect a node in an earlier layer to a node in a later layer.

[0145] In some embodiments, the neural network may be or include one or more recurrent neural networks. In some cases, at least some of the nodes of the recurrent neural network may form a cycle. Recurrent neural networks are particularly useful for processing inherently time-series input data. In particular, by utilizing recurrent or directed circular node connections, recurrent neural networks can transmit or retain information from preceding parts of an input data sequence to subsequent parts.

[0146] Some examples include time-series data (e.g., sensor data over time or images taken at different times). For example, a recurrent neural network can analyze sensor data over time to detect or predict swipe direction or perform handwriting recognition. Ordered input data includes words in a sentence (e.g., natural language processing, speech detection or processing), musical notes in a song, a sequence of user actions (e.g., sequential detection or prediction of application usage), a sequence of object states, etc. In some embodiments, recurrent neural networks include long-term memory (LSTM) recurrent neural networks, gated recurrent units, bidirectional recurrent neural networks, continuous-time recurrent neural networks, neural history compressors, echo state networks, Elman networks, Jordan networks, recurrent neural networks, Hopfield networks, fully recurrent networks, and inter-sequence transformation configurations.

[0147] In some examples, the neural network may be or include one or more non-recursive intersequence models, such as a self-attention-based transformer network. Details of an exemplary transformer network can be found at http: / / papers.nips.cc / paper / 7181-attention-is-all-you-need.pdf.

[0148] In some embodiments, a neural network may include one or more convolutional neural networks. In some cases, a convolutional neural network may include one or more convolutional layers that perform convolutions on input data using trained filters. Filters are also called kernels. Convolutional neural networks are particularly useful in visual problems where the input data includes images such as still images and videos. However, convolutional neural networks are also applicable to natural language processing.

[0149] In embodiments, the neural network may be or include one or more generative networks, such as a generative adversarial network. The generative network can be used to generate new data, such as new images or other content.

[0150] In some embodiments, a neural network may be an autoencoder or may include an autoencoder. In some cases, the purpose of an autoencoder is to learn a representation of a dataset (e.g., low-dimensional coding), usually for dimensionality reduction. For example, in some cases, an autoencoder attempts to encode input data and provide output data that reconstructs the input data from that encoding. In recent years, the concept of autoencoders has become more widely used to learn generative models for data. In some cases, an autoencoder may include additional loss in addition to reconstructing the input data.

[0151] In embodiments, the neural network may include one or more other forms of artificial neural networks, such as deep Boltzmann machines, deep belief networks, and stacked autoencoders. Any of the neural networks described herein can be combined (e.g., stacked) to form more complex networks.

[0152] Figure 4 shows an example of a neural network with multiple layers. The neural network 340 includes an input layer, a hidden layer, and an output layer, each consisting of multiple nodes or neurons that respond to different combinations of inputs from the previous layer. Numerical weights are assigned to the connections between neurons, determining the relative influence of the input on the output value of that node. The input layer may include multiple input nodes 342, 344, 346, 348, and 350 that provide information or input data from the external world (e.g., sensor data, image data, text data, audio data, etc.) to the neural network 340. The input data is obtained from different sources and may include library data x1, simulation data x2, user input data x3, training data x4, result data x5, etc. The input nodes 342, 344, 346, 348, and 350 only transmit information to the next layer and do not perform calculations themselves. The hidden layer may include multiple nodes such as nodes 352, 354, and 356. Hidden layer nodes 352, 354, and 356 process information from the input layer based on the weights of the connections between the input layer and the hidden layer, and transfer the information to the output layer. The output layer includes output node 358, which processes information based on the weights of the connections between the hidden layer and the output layer, and is responsible for computing and transferring information as output 359 from the network to the outside world, such as recognizing specific objects or activities or predicting states and actions.

[0153] In some embodiments, the neural network 340 includes two or more hidden layers and may be called a deep neural network. The layers are constructed such that the first layer detects primitive patterns in the input (e.g., image) data, the second layer detects patterns of patterns, and the third layer detects patterns of those patterns. In some embodiments, nodes in the neural network 340 may be connected to all nodes in the preceding and succeeding layers. These layers may therefore be called fully connected layers. In some embodiments, nodes in the neural network 340 may be connected to only some nodes in the preceding and succeeding layers. These layers may therefore be called loosely connected layers. Each neuron in the neural network consists of a weighted linear combination of its inputs, and the computation in each neural network layer can be described as the multiplication of the input matrix and the weight matrix. Next, a bias matrix is ​​added to the resulting product matrix, taking into account the threshold of each neuron in the next level. Furthermore, an activation function is applied to each resulting value, and the obtained values ​​are placed in the matrix of the next layer. Thus, the output from node i in the neural network is expressed as follows: yi = f (Σxiwi + bi) Here, f is the activation function, Σxiwi is the weighted sum of the input matrices, and bi is the bias matrix.

[0154] An activation function determines the level of activity or excitation generated at a node as a result of an input signal of a specific magnitude. The purpose of an activation function is to introduce nonlinearity into the output of a neural network node, because most real-world functions are nonlinear, and it is desirable for neurons to be able to learn these nonlinear representations. Multiple activation functions are available in artificial neural networks. One example is the sigmoid function σ(x), which is a continuous, S-shaped, monotonically increasing function that asymptotically converges to a fixed value as the input approaches positive or negative infinity. The sigmoid function σ(x) takes a real-valued input and converts it to a value between 0 and 1. σ(x) = 1 / (1 + exp(x))

[0155] Another activation function is the tangent hinge function, which converts real-valued inputs to values ​​in the range [?1,1]: tanh(x) = 2σ(2x)?

[0156] A third activation function is the rectified linear unit (ReLU) function. The ReLU function accepts real-valued inputs and thresholds values ​​greater than or equal to zero (i.e., it replaces negative values ​​with zero): ?(x)=max(0,x)

[0157] The activation functions described above are provided as examples, and in various embodiments, the neural network 340 includes (but is not limited to) combinations of activation functions such as identity function, binary step function, logistic function, soft step function, tan h function, arctan function, soft sine function, rectified linear unit (ReLU), leaky rectified linear unit, parametric rectified linear unit, randomized leaky rectified linear unit, exponential linear unit, S-shaped rectified linear activation unit, adaptive piecewise linear, soft plus, vent identity, soft exponential, sine wave, sinc, Gaussian, soft max, max out, and / or other activation functions.

[0158] In the example shown in Figure 4, input layer nodes 342, 344, 346, 348, and 350 receive external inputs x1, x2, x3, x4, and x5, which may be numerical depending on the input dataset. Although only five inputs are shown in Figure 4, it should be understood that in various implementations, nodes may contain tens, hundreds, thousands, or even more inputs. As mentioned earlier, no computation is performed in the input layer, so the outputs from input layer nodes 342, 344, 346, 348, and 350 are x1, x2, x3, x4, and x5, respectively, and these are input to the hidden layer. The output of hidden layer node 352 may depend on the outputs from the input layer (x1, x2, x3, x4, x5) and the weights associated with the connections (w1, w2, w3, w4, w5). Therefore, the output from node 352 is calculated as follows: Y 352==f(x1w1+x2w2+x3w3+x4w4+x5w5+b 352 )

[0159] The outputs from nodes 354 and 356 in the hidden layer are calculated in a similar manner and supplied to node 358 in the output layer. Node 358 in the output layer performs the same calculations as nodes 352, 354, and 356 in the hidden layer (using connection-related weights v1, v2, and v3). Y 358 =f(y 352 v1+y 354 v2+y 356 v3+b 358 ) Here Y 340 This represents the output of neural network 340.

[0160] As mentioned earlier, weights are associated with the connections between nodes in a neural network, which determine the relative influence of the input value on the output value of the target node. Before the network is trained, random values ​​are selected for each weight. The weights are adjusted during the training process, and this adjustment of weights to determine the optimal set of weights that maximizes the accuracy of the neural network is called training. For each input in the training dataset, the output of the artificial neural network is observed and compared to the expected output. The error between the expected output and the observed output is propagated to the previous layer, and the weights are adjusted based on this error. This process is repeated until the output error falls below a predetermined threshold.

[0161] In an embodiment, backpropagation (e.g., backpropagation of error) is used together with an optimization method such as gradient descent to adjust weights and update neural network characteristics. Backpropagation is a supervised learning method that learns from labeled training data and errors at nodes, and changes the parameters of the neural network to reduce errors. For example, the result of forward propagation (such as output activation values) determined using training input data is compared with the corresponding known reference output data, and the gradient of the loss function is calculated. This gradient is used in the optimization method to determine newly updated weights in an attempt to minimize the loss function. For example, to measure the error, the root mean square error is determined using the following equation. (eq.1) E = (target value - output value) 2

[0162] To determine the gradient with respect to the weight "w", the partial derivative of the error with respect to the weight is obtained as follows. (eq.2) Gradient = ∂E / ∂w

[0163] The calculation of the partial derivative of the error with respect to the weight can be propagated backward at the node level of the neural network. Then, a part of the gradient (e.g., ratio, percentage, etc.) is subtracted from the weight to determine the updated weight. This part is specified as the learning rate "a". Therefore, an example of the equation for determining the updated weight is as follows. (eq.3) w new = w old - α∂E / ∂w

[0164] The learning rate must be selected such that it is not too small (e.g., a rate that is too small may slow down convergence to the target weight) and not too large (e.g., a rate that is too large may cause the weight not to converge to the target weight).

[0165] After adjusting the weights, the network should exhibit better performance than before for the same input. This is because the weights have been adjusted to minimize the error.

[0166] As mentioned earlier, neural networks sometimes include convolutional neural networks (CNNs). CNNs are specialized neural networks designed to process data with known grid topologies, such as image data. Therefore, CNNs are widely used for classification, object recognition, and computer vision applications, but they can also be applied to other pattern recognition tasks such as speech and language processing.

[0167] Convolutional neural networks learn highly nonlinear mappings by interconnecting multiple layers of artificial neurons, each with an activation function that creates inter-layer dependencies. These typically include one or more convolutional layers, interspersed with one or more subsampling and nonlinear layers, and usually followed by one or more fully connected layers.

[0168] Referring to Figure 5, CNN360 comprises an input layer containing the input image 362 to be classified by CNN360, a hidden layer containing one or more convolutional layers (with one or more activation or nonlinear layers (e.g., ReLU), pooling layers, or subsampling layers scattered throughout), and an output layer (typically containing one or more fully connected layers). The input image 362 is represented by a matrix of pixels and may have multiple channels. For example, a color image may have red, green, and blue channels, each representing the red, green, and blue (RGB) components of the input image, respectively. Each channel is represented by a two-dimensional pixel matrix with pixel values ​​ranging from 0 to 255. On the other hand, a grayscale image may have only one channel. The following sections describe the processing of a single image channel using CNN360. It will be understood that multiple channels can be processed in a similar manner.

[0169] As shown in the figure, the input image 362 is processed by a hidden layer. The hidden layer consists of a set of convolutional layers 364 and activation layers 368, followed by pooling layers 366 and 370.

[0170] The convolutional layers of a convolutional neural network function as feature extractors that learn from an input image and decompose it into hierarchical features. The convolutional layer performs a convolution operation on the input image, during which a filter (also called a kernel or feature detector) slides across the input image in a specific step size (called the stride). At each position (or step), the element-wise multiplication of the filter matrix and the overlapping matrices in the input image is calculated and summed to obtain a final value representing a single element of the output matrix that constitutes the feature map. A feature map refers to image data that represents various features of the input image data and may have a smaller dimension compared to the input image. Activation layers (nonlinear layers) use different nonlinear trigger functions to clearly indicate the identification of features in each hidden layer. Nonlinear layers implement nonlinear triggers using various singular functions, such as the rectified linear unit (ReLU), hyperbolic tangent function, the absolute value of the hyperbolic tangent, and the sigmoid function. In some implementations, the ReLU activation function implements a function y=max(x,0), keeping the input and output sizes of the layer the same. The advantage of using ReLU is that it significantly speeds up the training of convolutional neural networks. ReLU is a discontinuous, non-saturating activation function that is linear when the input value is greater than 0, and 0 otherwise.

[0171] As shown in Figure 5, the first convolutional-activation layer 364 performs convolution on the input image 362 using multiple filters, and then applies a nonlinear operation (e.g., ReLU) to generate multiple output matrices (or feature maps) 372. The number of filters used is called the depth of the convolutional layer. Therefore, in the example in Figure 5, the first convolutional-activation layer 364 has a depth of 3 and generates three feature maps using three filters. The feature maps 372 are then passed to the first pooling layer, where a pooling function is used to subsample or downsample the feature maps to generate an output matrix 374. The pooling function reduces the number of parameters and computational complexity in the network by replacing the feature maps with summary statistics and reducing the spatial dimension of the extracted feature maps. Thus, the pooling layer reduces the dimensionality of the feature maps while retaining the most important information. The pooling function can also be used to introduce translation invariance into the neural network so that small translations to the input do not change the pooling output. Pooling layers can use different pooling functions, such as maximum pooling, average pooling, and 12-norm pooling.

[0172] The output matrix 374 is then processed by a second convolutional and activation layer 368, where convolution and nonlinear activation operations (e.g., ReLU) are performed as described above to generate a feature map 376. In the example shown in Figure 5, the second convolutional and activation layer 368 has a depth of 5 layers. The feature map 376 is passed to a pooling layer 370, where it is subsampled or downsampled to generate the output matrix 378.

[0173] The output matrix 378 generated by the pooling layer 370 is then processed by one or more fully connected layers 380 that constitute part of the output layer of the CNN 360. The fully connected layers 380 have full connectivity to all feature maps of the output matrix 378 of the pooling layer 370. In an embodiment, the fully connected layers 380 can take the output matrix 378 generated by the pooling layer 370 as input in vector form, perform higher-order decisions, and output feature vectors containing structural information in the input image. In an embodiment, the fully connected layers 380 can classify objects in the input image 362 into one of several categories using a softmax function. The softmax function is used as the activation function of the output layer, taking a vector of real-valued scores and mapping them to a vector of values ​​between 0 and 1 that sum to 1. In an embodiment, other classifiers, such as a support vector machine (SVM) classifier, may also be used.

[0174] In some embodiments, one or more normalization layers can be added to CNN360 to normalize the output of the convolutional filter. The normalization layers provide whitening or lateral suppression, avoid vanishing or exploding gradients, stabilize learning, and enable higher learning rates and faster convergence. In some embodiments, the normalization layer is added after the convolutional layer and before the activation layer.

[0175] Therefore, CNN360 can be viewed as a set of convolutional layers, activation layers, pooling layers, normalization layers, and fully connected layers stacked to learn, reinforce, and extract implicit features and patterns in the input image 362. As used herein, a “layer” refers to one or more components that operate in a similar function by mathematical or other functional means, processing the input they receive to produce / derive an output to the next layer containing one or more other components for further processing within CNN360.

[0176] The initial layers of CNN360 (e.g., convolutional layers) extract low-order features such as edges and gradients from the input image 362. Subsequent layers extract or detect increasingly complex features and patterns within the image data, such as curvature and the presence of texture. The output of each layer serves as input to subsequent layers within CNN360 to learn a hierarchical feature representation from the input image 362 data. This allows the convolutional neural network to efficiently learn increasingly complex and abstract visual concepts.

[0177] Although the example shows only two convolutional layers, this disclosure is not limited to the illustrative architecture, and the CNN360 architecture can consist of any number of layers in total, and any number of convolutional layers, activation layers, and pooling layers. For example, as described above, many variations and improvements have been made to the basic CNN model described above. Examples include AlexNet, GoogLeNet, VGGNet (a stack of many narrow convolutional layers and max pooling layers), residual networks or ResNet (which learn residual mapping using residual blocks and skip connections), DenseNet (which connects each layer of the CNN to all other layers in a feedforward manner), squeeze-excitation networks that incorporate global context into features, and amoeba networks that use evolutionary algorithms to explore and discover the optimal architecture for image recognition.

[0178] (Training a Convolutional Neural Network (CNN)) The learning process of convolutional neural networks such as CNN360 may be similar to the learning process described for neural network 340 in Figure 4.

[0179] In the embodiment, all parameters and weights (including weights in filters and weights in fully connected layers) are initially assigned (e.g., randomly). Then, during training, training images in which objects have been detected and classified are provided as input to CNN360, and CNN360 performs a forward propagation step. That is, CNN360 applies convolution, nonlinear activation, and pooling layers to each training image to determine classification vectors (i.e., detection and classification of each training image). These classification vectors are compared to predefined classification vectors. The error between the CNN's classification vectors and the predefined classification vectors (e.g., difference of sum of squares, log loss, softmax log loss) is calculated. This error may be used to update the CNN's weights and parameters in a backpropagation process involving one or more iterations, possibly using gradient descent. This training process is repeated for each training image in the training set.

[0180] The training and inference processes described above can be performed using hardware, software, or a combination of hardware and software. However, training a convolutional neural network like CNN360, or performing inference using a trained CNN, generally requires enormous computing power to perform operations such as matrix multiplication and convolution. Therefore, dedicated hardware circuits such as graphics processing units (GPUs), tensor processing units (TPUs), neural network processing units (NPUs), FPGAs, ASICs, and other highly parallelized processing circuits can be used for training and / or inference. Training and inference are performed in the cloud, on data centers, or on devices.

[0181] (Region-based CNN (RCNN) and object detection) In some embodiments, the object detection model extends the capabilities of CNN-based image classification neural network models by not only classifying objects but also determining their location within an image using bounding boxes. The region-based CNN (R-CNN) method is used to extract regions of interest (ROIs), each of which is a rectangle that may represent the boundary of an object in the image. Conceptually, R-CNN operates in two stages. In the first stage, the region proposal method generates all potential bounding box candidates in the image. In the second stage, a CNN classifier is applied to each proposal to identify the object. Alternatively, a fast R-CNN architecture can be used, which is a single network integrating the feature extractor and classifier. Even faster R-CNNs exist, which integrate the region proposal network (RPN) and fast R-CNN into an end-to-end trainable framework. Mask R-CNN adds instance segmentation capabilities, and Mesh R-CNN adds the ability to generate 3D meshes from 2D images.

[0182] Referring to Figure 3, in an embodiment, the artificial intelligence module 304 can provide access to and / or integrate with the robotic process automation (RPA) module 316. The RPA module 316 can, among other things, facilitate computer automation of workflow creation and validation. The RPA module 316 provides automation of tasks that are performed by humans. For example, receiving and confirming written information, entering data into user interfaces, transforming and processing data such as files and records, recording observation results, generating documents such as reports, and communicating with other users by means such as email. In some cases, tasks involve workflows that include multiple interrelated steps, contextual information related to the task, and interactions with other applications or humans. The RPA module 316 can be configured to receive or learn one or more such workflows on behalf of humans, and in a manner similar to human behavior and logic, and then execute such workflows in response to various triggers such as events. Examples of the RPA module 316 include those described in this disclosure and the literature incorporated herein, and may involve the automation of any of the broad value chain network activities or entities described therein.

[0183] In embodiments, the RPA module 316 is configured to receive or learn robotic process automation workflows in various ways. As a first example, in embodiments, the RPA module 316 may include a graphical user interface (GUI) that allows a user to specify details of a robotic process automation workflow. The GUI may include components that represent different types of actions, such as input receiving actions from a user or application, data transformation or other processing actions, and input providing actions to an application. The GUI may receive from the user a selection of components that represent actions corresponding to steps in a workflow when performed by a human. The GUI may also receive from the user the interconnections between the selected components (e.g., the logical order in which corresponding actions are performed, or relationships in which one component depends on another (e.g., data output by one component is received as input by another)). The GUI may include one or more templates, such as a set of actions that are performed together to complete a common workflow. The GUI may receive from the user a selection of templates, which may also include one or more details to adapt the selected templates to a specific human workflow. Based on the input received from the user, the RPA module 316 may generate a robotic process automation workflow that is executable to perform the workflow. RPA module 316 can save the generated workflow for future use. For example, RPA module 316 can execute the compiled code or interpret the generated script to run the workflow in the same way a human would.

[0184] As a second example, in this embodiment, the RPA module 316 is configured to receive or learn a set of rule-based workflows. For example, the RPA module 316 may include a GUI that allows a user to specify the details of a robotic process automation workflow as a set of conditions and corresponding actions. The GUI may include a set of components that respond to monitored conditions, such as the state of a resource or the occurrence of an event. The GUI for designing the workflow may include a set of components that represent the corresponding actions when any of the conditions occur. The GUI may receive from the user a selection of components that represent one or more conditions of the workflow, and a selection of one or more components that represent the actions to be performed in response to the conditions. In some embodiments, the GUI may include one or more templates, such as one or more conditions associated with one or more actions corresponding to a common workflow. The GUI may receive from the user a selection of templates, including one or more details that adapt the selected templates to a specific workflow performed by a human. Based on the input received from the user, the RPA module 316 can generate a robotic process automation workflow that automates a set of tasks in response to one or more detected events. The RPA module 316 may save the generated workflow for future use. For example, RPA module 316 can monitor selected conditions and, in response to the occurrence of the selected action, execute the selected action in a manner similar to that a human would perform.

[0185] As a third example, in an embodiment, the RPA module 316 is configured to learn a workflow by recording a series of actions performed by a human to complete the workflow. For example, the RPA module 316 can receive instructions from a user indicating the start of a workflow involving a device (e.g., selecting the “Start Recording” button). The RPA module 316 can receive user input from the user, such as input to one or more human interaction devices (HIDs) such as a keyboard, mouse, touchscreen, camera, or microphone. Alternatively, the RPA module 316 can receive user input as a series of human interaction events reported by a device (e.g., an operating system input layer that receives and aggregates user input from one or more human input devices). Alternatively, the RPA module 316 can receive user input as a series of events reported by one or more applications (e.g., a web browser that reports a series of user input events). The RPA module 316 can record user input as an input sequence. The RPA module 316 can associate the recorded user input with contextual information, such as identifying the application to which the input was directed. The RPA module 316 can associate recorded user input with other events (e.g., preliminary events of the application receiving user input (e.g., notification by the web browser that the web page has finished rendering and is ready to accept user input) and / or application response events to the receipt of user input (e.g., actions taken by the web page in response to the receipt of user input)). The RPA module 316 can associate recorded user input with other events occurring within the device (e.g., actions taken by other applications or the device's operating system in response to user input). The RPA module 316 can receive instructions from the user to end the workflow (e.g., selection of the stop recording button).The RPA module 316 can generate a workflow that includes a record of observed user input and can be associated with other data as needed. The RPA module 316 can save the generated workflow for future use. For example, the RPA module 316 can replay the recorded sequence of user input and execute the workflow in a similar manner to how a human would have done it.

[0186] As a fourth example, in the embodiment, the RPA module 316 is configured to learn workflows by observing human-device interactions. For example, a human may use the device to perform multiple workflows over a period of time, such as one business day. The RPA module 316 can monitor human user input and identify one or more behavioral patterns that the human repeatedly performs from that input. The RPA module 316 can determine that the behavioral patterns correspond to workflows performed by the human. In some embodiments, the RPA module 316 can identify differences between different instances of behavior performed by the human during the workflow (e.g., different types of data inputs occurring in different instances of behavior). The RPA module 316 can associate behaviors in the workflow with one or more parameters, where the parameters correspond to different variations between different instances of behavior performed by the human. In various embodiments, the RPA module 316 can determine the rationale for each variation of behavior associated with different variations of behavior in the workflow. For example, the RPA module 316 can determine that if the workflow is performed by a human on behalf of a first user, the action should be performed using a first data input value (e.g., a data input containing a first username). When a workflow is executed by a human on behalf of a second user, the actions should be performed using a second data entry value (e.g., a data entry containing the second user's username). The data entry is represented within the workflow as a data entry parameter (e.g., the name of the user on whom the workflow is executed) and, if necessary, accompanied by a specific value corresponding to the workflow context (e.g., the name of the user on whom the workflow can be executed). RPA module 316 can generate a workflow containing a set of commands corresponding to patterns of operations performed by the user during the workflow, and can optionally include parameters and / or parameter values ​​for various operations in the workflow. RPA module 316 can save the generated workflow for future use.For example, the RPA module 316 can replay a sequence of commands to replicate the operation patterns corresponding to a workflow, as if they were executed in the same way as human operations.

[0187] In some embodiments, the RPA module 316 can be implemented in various architectures. As a first example, the RPA module 316 can be implemented on a device used by a human to execute a workflow and / or on a device used by a user to specify workflow details. The RPA module 316 can store one or more generated workflows on the device and execute the workflows on the same device. As a second example, the RPA module 316 can be implemented on the first device to replicate a workflow performed by a human on a second device. The RPA module 316 can monitor the operations performed by the human on the second device when executing a task, generate and save the workflow on the first device, and execute that workflow on the first device to execute the task on the first device in the same way that the user did on the second device. As a third example, the RPA module 316 is implemented on the first device to generate a workflow corresponding to a task performed by a human on the first device and send it to the second device. This workflow causes the second device to execute the task in the same way that the user did on the first device. As a fourth example, the RPA module 316 is implemented on a second device and can receive workflows corresponding to tasks performed by a human on the first device. The workflow of the RPA module 316 executes on the second device, and the user performs tasks on the second device in the same way as they did on the first device. In some embodiments, the RPA module 316 can be distributed across a group of two or more devices. For example, a first part of the RPA module 316 runs on the first device and generates workflows based on the interaction between the human and the first device, and a second part of the RPA module 316 runs on the second device and executes the workflow on the second device. In some embodiments, at least a portion of the RPA module 316 can be replicated on multiple devices (e.g., two or more devices that execute workflows generated based on the interaction between the human and the first device (e.g., in parallel and / or sequentially)).In some embodiments, different RPA modules 316 running on multiple devices can work together to execute one or more workflows (e.g., a first RPA module 316 running on a first device to execute the first part of a workflow, and a second RPA module 316 running on a second device to execute the second part of the same workflow). Each RPA module 316 may operate in a specific role when executing at least part of a workflow. For example, a first RPA module 316 running on a cloud edge device to receive workflow inputs, a second RPA module 316 running on a cloud server to process workflow inputs, and a third RPA module 316 running on another cloud edge device to present workflow outputs.

[0188] In this embodiment, the RPA module 316 can execute workflows in response to various triggers. The RPA module 316 can execute workflows in response to user requests, such as requests to execute code to run a learned workflow or requests to execute a specific script. The RPA module 316 can execute workflows in response to the detection of human activity patterns (e.g., a second workflow executed by the RPA module 316 in response to the completion of an initial workflow by a human). The RPA module 316 can execute at least a portion of a workflow instead of a human executing at least a portion of it. For example, the RPA module 316 can detect the start of a workflow by a human and propose to the human that the RPA module 316 execute the rest of the workflow. Upon receiving acceptance of the proposal, the RPA module 316 can execute the entire workflow on behalf of the human and / or execute one or more remaining steps of the workflow following the initial steps executed by the human. The RPA module 316 can execute workflows in response to the occurrence of specific data types (e.g., when a device receives a file containing a specific data type, such as a specific type of document or image). The RPA module 316 can execute workflows in response to messages received via communication channels such as email, phone calls, text messages, gesture input received from cameras and haptic input devices, and voice input received from microphones. The RPA module 316 can execute workflows in response to requests from operating systems or applications running on the device (e.g., requests from a spreadsheet application when a user enters a specific data type). The RPA module 316 can execute workflows in response to detected events. For example, if the device detects the presence of a particular person (e.g., if the device's camera detects a person's face), the RPA module 316 can execute a workflow that displays a report for that person. The RPA module 316 can execute workflows at scheduled intervals, such as every hour or every day.The RPA module 316 can execute a workflow in response to a request received from another workflow running on the same device or a different device (e.g., a second workflow that runs after the first workflow is completed).

[0189] In this embodiment, the RPA module 316 can execute a workflow based on various inputs. The RPA module 316 can execute a workflow based on one or more details about the trigger of the workflow. For example, if the workflow is executed in response to a request from a user to execute the workflow, the RPA module 316 can execute the workflow based on one or more details about that request. For example, if the workflow is triggered by a user request to process a specific document, the RPA module 316 can execute the workflow based on one or more details about that document. If the workflow is executed as a message or phone response, the RPA module 316 can execute the workflow based on the identity of the message sender or caller. If the workflow is executed as a daily instance based on a schedule, the RPA module 316 can execute the workflow based on the day of the week on which the workflow is executed. If the workflow is executed in response to the detection of a condition, the RPA module 316 can execute the workflow based on one or more details of the condition. For example, if the condition is that the device's storage capacity exceeds a storage capacity threshold, the RPA module 316 can execute the workflow based on the severity of the storage capacity state (e.g., the device's remaining storage capacity). The RPA module 316 can execute workflows based on data sources such as one or more files in a file system, one or more rows or records in a database, or one or more messages received by a network interface. If the RPA module 316 is executing a workflow in response to one or more events, it can execute the workflow based on one or more details of the events. For example, if the RPA module 316 is executing a second workflow in response to the completion of a first workflow on the same or a different device, the RPA module 316 can execute the workflow based on the completion date and time of the first workflow, the results of the first workflow, and / or the output of the first workflow.RPA module 316 can execute workflows based on one or more contextual details. For example, RPA module 316 can execute workflows based on the number and identity of detected humans present near the device. RPA module 316 can execute workflows based on data associated with applications running on the device. For example, if RPA module 316 executes a workflow based on web page loading, RPA module 316 can execute a workflow based on data scraped from the web page content. RPA module 316 can execute workflows based on observations of human behavior involving interactions with hardware elements, software interfaces, and other elements. Observations include on-site observations of humans performing actual tasks, as well as observations in simulations and other activities. In the latter, humans perform actions with the explicit intention of providing training datasets and inputs for RPA module 316. For example, humans may characterize and label training datasets to help RPA module 316 learn to recognize and classify features and objects (there are many other examples).

[0190] In this embodiment, the RPA module 316 can interact with one or more applications while executing a workflow. For example, the RPA module 316 can extract data from application variables or objects, such as text content in a text box within a web form or cell contents in a spreadsheet. The RPA module 316 can extract data stored within the application (e.g., by inspecting the application's memory space). The RPA module 316 can analyze data generated as output by the application (e.g., one or more files generated by the application, one or more rows or records in a spreadsheet generated by the application, one or more network communication messages received and / or sent by the application over a network). The RPA module 316 can request data from the application by calling the application's application programming interface (API), and can receive and analyze data provided by the application in response to the API call. The RPA module 316 can inspect one or more properties of the device on which the application is running (e.g., the display area of ​​the device containing the application's graphical user interface) and extract data from the application. Alternatively, or additionally, the RPA module 316 can provide data to the application or modify the application's behavior while the workflow is running. For example, RPA module 316 can generate user input directed to an application (e.g., simulate a human interaction device (HID) such as a keyboard and generate keystrokes that are sent to the application as user input). RPA module 316 can directly send and / or modify data in an application (e.g., modify HTML data stored in a rendered web page to change the contents of a text box, or directly modify data in the application's memory space).RPA module 316 can request the operating system to interact with an application and / or change the application's behavior (e.g., request the device to start, activate, suspend, resume, terminate, or force quit an application). RPA module 316 can call an application's API to provide data to the application (e.g., call a spreadsheet API to request data entry into a specific cell). RPA module 316 can call code associated with an application to provide data or change the application's behavior (e.g., execute code encoded in an application-specific programming language and embedded in a document used by the application, or call a stored procedure in a database associated with the application). RPA module 316 can make the interaction with the application visible to humans (e.g., RPA module 316 provides user input that mimics a user visually launching a spreadsheet application and visually entering data into its various cells). RPA module 316 can hide the interaction with the application from humans (e.g., visually hide the application window while data is being entered into one or more text boxes within the application window).

[0191] In embodiments, the RPA module 316 can utilize various logical processes when executing a workflow. The RPA module 316 can acquire, interpret, analyze, transform, validate, aggregate, split, render, store, and / or otherwise process data associated with or received by a workflow. The RPA module 316 can send data to another workflow, application, or device for processing or storage, and / or query or receive data from another workflow, application, or device. The RPA module 316 can apply optical character recognition (OCR) processes to images (e.g., photographs of forms or documents) to identify and extract text content from images. The RPA module 316 can apply computer vision processing to images (e.g., photographs taken with a camera) to identify and extract image data from images. This includes detecting, recognizing, classifying, and / or locating one or more objects. The RPA module 316 can apply speech recognition processing to voice input (e.g., phone calls or voice input from a microphone) to identify and extract voice content (e.g., one or more voice commands). The RPA module 316 can apply a gesture recognition process to an input device (e.g., a camera that detects hand movements, a proximity sensor, or an inertial measurement unit) to identify one or more gestures performed by a human. The RPA module 316 can also apply a pattern recognition process to data to detect one or more patterns within the data (e.g., analyzing sensor data from a machine to detect the occurrence of one or more events related to the machine (e.g., the movement of the machine's moving parts)).

[0192] In this embodiment, the RPA module 316 executes a workflow in cooperation with a human or another workflow. For example, a workflow may include one or more human parts executed by a human and one or more automated parts executed by the RPA module 316. The RPA module 316 can first execute the automated parts and provide the results to the human, so that the human can execute the human parts based on those results. The RPA module 316 can receive the results of the human parts of the workflow and execute the automated parts of the workflow on those results. While the human is executing the human parts of the workflow, the RPA module 316 can execute the automated parts of the workflow in parallel and then integrate the results of the automated parts and the human parts of the workflow. The RPA module 316 can execute the first automated part of the workflow, present the results for human review and verification, and then execute a second automated part of the workflow based on the review and verification of the results of the first automated part.

[0193] In embodiments, the RPA module 316 can learn to perform specific tasks based on learned patterns and processes. The RPA module 316 can use one or more artificial intelligence modules 304 to perform one or more steps in a workflow. For example, the RPA module 316 can perform a data classification step on input data by applying a classification neural network to the input data. The RPA module 316 can perform a pattern recognition step on input data by applying a pattern recognition neural network to the input data. The RPA module 316 can perform a computer vision processing step and / or optical character recognition step in a workflow by applying one or more CNN(360) to an image. The RPA module 316 can perform a sequential analysis step involving time-series data by applying one or more recurrent neural networks (RNN) to the time-series data. The RPA module 316 can perform one or more natural language processing steps by applying one or more transformer-based neural networks to a natural language representation (e.g., a natural language document or natural language speech input).

[0194] In various embodiments, the RPA module 316 uses one or more untrained artificial intelligence modules 304. For example, one or more artificial intelligence modules 304 may include a k-nearest neighbor model that determines the classification of an incoming input based on its proximity to a set of other inputs that have known classifications of the incoming input. The k-nearest neighbor model classifies the incoming input according to a majority vote of the known classifications of the k inputs closest to it.

[0195] In various embodiments, the RPA module 316 uses one or more artificial intelligence modules 304 trained in an unsupervised learning manner. For example, the workflow may include an anomaly detection step, such as identifying a portion of a form containing handwritten text. The anomaly detection algorithm can divide the form into sets of symbols and compare the symbols to distinguish between frequently occurring symbols (e.g., machine-printed characters in a font) and less frequently occurring symbols (e.g., unique or at least highly variable handwritten characters). Thus, the anomaly detection algorithm can divide the form into areas containing machine-printed characters and areas containing handwritten characters. The RPA module 316 then processes each document area with either an OCR module configured to recognize machine-printed characters in a font, or an OCR module configured to recognize handwritten characters.

[0196] In various embodiments, the RPA module 316 utilizes one or more artificial intelligence modules 304 that are specifically designed and / or trained for a workflow. For example, the workflow may be associated with a training dataset, and the RPA module 316 may train one or more machine learning models to perform the workflow processing based on the training dataset. In various embodiments, the RPA module 316 uses one or more pre-trained artificial intelligence modules 304 to perform the workflow processing. For example, the RPA module 316 may receive a partially pre-trained natural language processing (NLP) machine learning model that is generally trained to recognize sentence structure and word meanings. The RPA module 316 may adapt the partially pre-trained NLP machine learning model based on the natural language expressions specifically associated with the workflow. This adaptation may include applying transfer learning to the artificial intelligence module 304 (e.g., training one or more classification layers of the classification portion of the NLP machine learning model more specifically while keeping other parts of the NLP machine learning model constant). The adaptive process may include retraining the artificial intelligence module 304 (e.g., retraining the entire NLP machine learning model based on the natural language expressions related to the workflow). Adaptation involves generating an ensemble of artificial intelligence modules 304 to execute a workflow (e.g., two or more artificial intelligence modules 304 each classify data in different ways, and the workflow's output classification is based on the agreement of the two or more artificial intelligence modules 304). An artificial intelligence module 304 may include a random forest in which one or more decision trees each analyze the input data according to different criteria, and the output of the random forest is determined based on the agreement of the decision trees. An artificial intelligence module 304 may also include a stacking ensemble in which two or more machine learning models each process data and produce an output, and another machine learning model determines which output from among the outputs of the two or more machine learning models to use as the output for data processing.

[0197] In embodiments, the RPA module 316 generates one or more outputs or results of a workflow. The RPA module 316 can generate data that can be stored by a device (e.g., as a file in a file system, or as a row or record in a database) as output. The RPA module 316 can generate data that is contained in another dataset (e.g., text entered in a form field, a number entered in a spreadsheet cell, or text entered in a text box on a web page) as output. The RPA module 316 can generate data that is sent to another device (e.g., sending form data from a web page to a web server) as output. The RPA module 316 can generate data that is communicated to one or more users as output (e.g., a visual notification of the result displayed to the user on the device, or a message sent to the user via a communication channel such as email, text message, or voice output). The RPA module 316 can generate data that modifies the behavior of an application (e.g., a command instructing the application to start, activate, pause, resume, stop, or terminate). RPA module 316 can generate data as output that modifies the operation of a device or other devices (e.g., commands to control machines such as printers, cameras, devices, and industrial manufacturing equipment). RPA module 316 can generate data as output that reflects the initial, current, or final state of a workflow (e.g., a dashboard showing the progress towards the completion of a workflow, or a representation of the workflow results in combination with the results of other workflows). RPA module 316 can generate one or more events as output as notifications about the progress, completion, and / or results of a workflow (e.g., notifications to humans, applications, device operating systems, or other devices). Events are received and further processed by RPA module 316 or another RPA module running on the same or a different device.For example, once the first workflow is complete, the RPA module 316 can start a second workflow based on the results and / or output of the first workflow. The RPA module 316 can generate output that documents one or more results of the workflow. For example, the RPA module 316 can update the log to document the workflow results and / or output, including one or more errors, exceptions, or validation failures that occurred during the workflow.

[0198] In this embodiment, the RPA module 316 modifies the workflow based on its performance. For example, the RPA module 316 may request a user review of one or more outcomes of the workflow, including one or more errors, exceptions, or validation failures that occurred during the workflow. The RPA module 316 may disable one or more steps or modules of the workflow that caused the errors, exceptions, or validation failures. The RPA module 316 may automatically adjust the workflow to run future instances of the workflow based on completed instances of the workflow. For example, the RPA module 316 may update the workflow to improve its efficiency, add or remove functionality from the workflow, adjust the functionality of the workflow to different behavior, log one or more instances and / or parameters of the workflow, and / or eliminate or mitigate one or more logical defects in the workflow. The RPA module 316 may update one or more artificial intelligence modules 304 associated with the workflow. For example, the RPA module 316 may generate or add one or more machine learning models to the workflow to improve its processing. The RPA module 316 may remove one or more machine learning models to improve the efficiency of the workflow. RPA module 316 can redesign and / or retrain one or more machine learning models based on workflow results. RPA module 316 can add one or more machine learning models to an existing ensemble of machine learning models.

[0199] (Analytics module) In some embodiments, the artificial intelligence module 304 may include and / or provide access to the analysis module 318. In some embodiments, the analysis module 318 is configured to perform various analytical processes on data output from value chain entities or other data sources. In exemplary embodiments, the analysis generated by the analysis module 318 facilitates the quantification of system performance against a set of goals and / or metrics. These goals and / or metrics may be pre-configured or dynamically determined from operational results. Examples of analytical processes that can be performed by the analysis module 318 are discussed below and in documents incorporated herein by reference. In some embodiments, analytical processes may include tracking goals or specific metrics with the coordination of value chain activities and demand intelligence, including demand forecasting for relevant item sets based on location, time, etc. (among many others).

[0200] (Digital Twin Module) In an embodiment, the artificial intelligence module 304 may include and / or provide access to the digital twin module 320. The digital twin module 320 may encompass any of the broad range of functions and capabilities described herein. In an embodiment, the digital twin module 320 may be configured to provide an execution environment for various types of digital twins, including, among other things, a twin of a physical environment, a twin of a robotic operating unit, a logistics twin, a management digital twin, an organizational digital twin, and a role-based digital twin. In an embodiment, the digital twin module 320 may be configured according to the digital twin systems and / or modules described in other parts of this disclosure. In an exemplary embodiment, the digital twin module 320 may be configured to generate a digital twin requested by an intelligence client 336. Furthermore, the digital twin module 320 may be configured to include an interface, such as an API, for receiving information from an external data source. For example, the digital twin module 320 may receive real-time data from sensor systems of machines, vehicles, robots, and other devices, and / or sensor systems of the physical environment in which the devices operate. In embodiments, the digital twin module 320 can receive digital twin data from other suitable data sources, such as third-party services (e.g., weather services, traffic data services, logistics systems, and databases). In embodiments, the digital twin module 320 may include digital twin data representing the characteristics, status, etc., of supply chain infrastructure entities, transportation or logistics entities, value chain network entities such as containers and cargo, as well as demand entities such as customers, merchants, stores, sales locations, and usage locations.The digital twin module 320 can be integrated with, linked to, or interact with an interface (e.g., a control tower or dashboard) for supply and demand coordination (including coordination of automation within supply chain activities and demand management activities).

[0201] In one embodiment, the digital twin module 320 can provide and manage access to a digital twin library. The artificial intelligence module 304 can access the library to perform functions such as simulating behavior in a given environment in response to specific stimuli.

[0202] (Machine vision module) In some embodiments, the artificial intelligence module 304 may include and / or provide access to the machine vision module 322. In some embodiments, the machine vision module 322 is configured to process images (e.g., those captured by a camera) to detect and classify objects within the images. In some embodiments, the machine vision module 322 receives one or more images (which may be frames from a video feed or a single still image) and identifies “blobs” within the images (e.g., using edge detection techniques). The machine vision module 322 then classifies the blobs. In some embodiments, the machine vision module 322 leverages one or more machine learning image classification models and / or neural networks (e.g., convolutional neural networks) to classify the blobs in the images. In some embodiments, the machine vision module 322 performs feature extraction on the images and / or each blob within the images before classification. In some embodiments, the machine vision module 322 leverages the classification performed on the previous images to confirm or update the classification from the previous images. For example, if an object detected in a previous frame was classified with a low confidence score (e.g., the object was partially occluded or out of focus), the machine vision module 322 can confirm or update the classification if it can determine the object's classification with higher confidence. In embodiments, the machine vision module 322 is configured to detect occlusions, such as objects that may be occluded by another object. In embodiments, the machine vision module 322 receives additional inputs to assist in the image classification task from radar, sonar, a digital twin of the environment (which may show the locations of known objects), and / or similar. In some embodiments, the machine vision module 322 includes or interfaces with a liquid lens.In these embodiments, liquid lenses can facilitate improvements in machine vision (e.g., when the environment or robot's work requires focusing on multiple distances) and / or other machine vision tasks made possible by liquid lenses.

[0203] (Natural Language Processing Module) In an embodiment, the artificial intelligence module 304 may include and / or provide access to a natural language processing (NLP) module 324. In an embodiment, the NLP module 324 performs natural language tasks on behalf of the intelligence service client 336. Examples of natural language processing techniques include, but are not limited to, speech recognition, speech segmentation, speaker diarization, text-to-speech conversion, lemmatization, morphological segmentation, part-of-speech tagging, stemming, syntactic analysis, and lexical analysis. In an embodiment, the NLP module 324 enables voice commands received from a human. In an embodiment, the NLP module 324 may receive an audio stream (e.g., from a microphone) and perform speech-to-text conversion on the audio stream to obtain a transcript of the audio stream. The NLP module 324 may use various NLP techniques (e.g., NLP models, neural networks, and / or similar) to process the text (e.g., the transcript of the audio stream) and determine the meaning of the text. In one embodiment, the NLP module 324 can determine an action or command spoken in the audio stream based on the NLP results. In another embodiment, the NLP module 324 can output the NLP results to the intelligence service client 336.

[0204] In some embodiments, the NLP module 324 provides the intelligence service client 336 with the ability to analyze one or more conversational voice instructions provided by a human user to perform one or more tasks, and the ability to communicate with the human user. The NLP module 324 can perform speech recognition to recognize voice instructions, perform natural language understanding to analyze the instructions and derive their meaning, and perform natural language generation to process user instructions and generate voice responses for the user. In some embodiments, the NLP module 324 enables the intelligence service client 336 to understand the instructions and provides a response to the user upon successful completion of the task by the intelligence service client 336. In embodiments, the NLP module 324 can construct questions to ask the user if the context of the user request is not entirely clear. In embodiments, the NLP module 324 can utilize input from one or more sensors, including a visual sensor and location data (e.g., GPS data), to determine contextual information related to the processed voice or text data.

[0205] In the embodiment, the NLP module 324 uses neural networks, including recurrent neural networks, long short-term memory (LSTM), gated recurrent units (GRUs), transformer neural networks, and convolutional neural networks, when performing NLP tasks.

[0206] Figure 6 shows an exemplary neural network for implementing the NLP module 324. In the illustrated example, the exemplary neural network is a transformer neural network. In this example, the transformer neural network includes three input stages and five output stages to transform an input sequence into an output sequence. This transformer example includes an encoder 382 and a decoder 384. The encoder 382 processes the input, and the decoder 384 generates output probabilities. The encoder 382 has three stages, and the decoder 384 has five stages. Stage 1 of the encoder 382 represents the input as a sequence of position encodings added to the embedded input. Stages 2 and 3 of the encoder 382 include N layers (e.g., N=6), each layer containing a position-specific feedforward neural network (FNN) and an attention-based sublayer. Each attention-based sublayer in stage 2 of the encoder 382 contains four linear projections and multi-head attention logic, which are added and normalized and fed into the position-specific FNN in stage 3 of the encoder 382. The second and third stages of encoder 382 employ residual connection and subsequent normalization layers at the output.

[0207] The exemplary decoder 384 processes the output embedding as its input. The output embedding is right-shifted by one position to help ensure that the prediction for position i depends on positions prior to i. In stage 2 of decoder 384, masked multi-head attention is corrected to prevent a position from directing attention to subsequent positions. Stages 3-4 of decoder 384 contain N layers (e.g., N=6), each layer consisting of a position-by-position FNN and two attention-based sublayers. Each attention-based sublayer in stage 3 of decoder 384 contains four linear projections and multi-head attention logic, which are added and normalized and fed into the position-by-position FNN in stage 4 of decoder 384. Stages 2-4 of decoder 384 employ residual connections and subsequent normalization layers at the output. Stage 5 of decoder 384 provides a linear transformation and subsequent softmax function to normalize the resulting vector of K numbers into a probability distribution containing K probabilities proportional to an exponential function of the K input numbers.

[0208] Further examples of neural networks are described elsewhere in this specification.

[0209] (Rule-based module) Referring to Figure 3, in some embodiments, the artificial intelligence module 304 may include and / or provide access to a rule-based module 328 that is integrated with or accessible from the intelligence service client 336. In some embodiments, the rule-based module 328 consists of program logic that defines a set of rules and other conditions that trigger specific actions performed in relation to the intelligence client. In some embodiments, the rule-based module 328 may consist of program logic that receives input and determines whether one or more rules are met based on that input. If the conditions are met, the rule-based module 328 determines what action to take, which may be output to the requesting intelligence service client 336. The data received by the rule-based engine may be received from the intelligence service input 332 source or requested from other modules within the artificial intelligence module 304 (e.g., machine vision module 322, neural network module 314, machine learning module 312, etc.). For example, the rule-based module 328 receives classification information of objects within the field of view of a mobile system (such as a robot or autonomous vehicle) from a machine vision system and / or sensor data from the mobile system's LiDAR sensors, and accordingly determines whether the mobile system should continue its path, change course, or stop. In embodiments, the rule-based module 328 may be configured to make other appropriate rule-based decisions on behalf of each client 336. Examples of such decisions are discussed throughout this specification. In some embodiments, the rule-based engine may apply governance criteria and / or analysis modules, which are described in more detail below.

[0210] (Intelligence Service Controller and Analytics Management Module) In this embodiment, the artificial intelligence module 304 interfaces with the intelligence service controller 302. The intelligence service controller 302 is configured to determine the type of request issued by the intelligence service client 336, and accordingly can determine the set of governance criteria and / or analyses that the artificial intelligence module 304 should apply when responding to the request. In this embodiment, the intelligence service controller 302 may include an analysis management module 306, a group of analysis modules 308, and a governance library 310.

[0211] In the embodiment, the intelligence service controller 302 is configured to determine the type of request issued by the intelligence service client 336, and accordingly can determine a set of governance criteria and / or analyses to be applied by the artificial intelligence module 304 when responding to the request. In the embodiment, the intelligence service controller 302 may include an analysis management module 306, a group of analysis modules 308, and a governance library 310. In the embodiment, the analysis management module 306 receives a request from the artificial intelligence module 304 and determines the governance criteria and / or analyses to which the request is involved. In the embodiment, the analysis management module 306 can determine the governance criteria to be applied to the request based on the type of decision requested and / or whether a particular analysis is performed with respect to the requested decision. For example, a request for a control decision resulting from an action by the intelligence service client 336 may imply a specific governance criterion to be applied (such as a safety criterion, a legal criterion, or a quality criterion) and / or imply one or more analyses related to the control decision, such as a risk analysis, a safety analysis, or an engineering analysis.

[0212] In some embodiments, the analysis management module 306 can determine governance criteria applicable to a decision request based on one or more conditions. Non-limiting examples of such conditions include the type of decision requested, the geographical location where the decision is made, the environment in which the decision will affect, and the current or projected environmental conditions of the environment. In embodiments, governance criteria may be defined as a set of criteria libraries stored in the governance library 310. In embodiments, a criteria library may define conditions, thresholds, rules, recommendations, or other appropriate parameters for analyzing a decision. Examples of standard libraries include legal standard libraries, regulatory standard libraries, quality standard libraries, technical standard libraries, safety standard libraries, financial standard libraries, and / or other appropriate types of standard libraries. In embodiments, the governance library 310 may include an index that indexes specific standards defined in each standard library based on different conditions. Examples of conditions include the jurisdiction or geographical area to which a particular standard applies, the environmental conditions to which a particular standard applies, the device type to which a particular standard applies, the material or product to which a particular standard applies, and / or similar.

[0213] In some embodiments, the analytics management module 306 may determine an appropriate set of standards to apply to a particular decision and provide that appropriate set of standards to the artificial intelligence module 304. This allows the artificial intelligence module 304 to leverage relevant governance standards when making decisions. In these embodiments, the artificial intelligence module 304 may be configured to apply the standards in the decision-making process. This ensures that the decisions output by the artificial intelligence module 304 are consistent with the associated governance standards. It is understood that the standards library within the governance library may be defined by the platform provider, customers, and / or third parties. The standards may be government standards, industry standards, customer standards, or other appropriate sources. In embodiments, each set of standards may include a set of conditions used to determine which standards should be applied in the context, thereby ensuring that each set of standards is applied.

[0214] In some embodiments, the analysis management module 306 may determine one or more analyses to be performed with respect to a particular decision and provide the artificial intelligence module 304 with corresponding analysis modules 308 to perform those analyses. This allows the artificial intelligence module 304 to leverage the corresponding analysis modules 308 to analyze the decision before outputting the decision to the requesting client. In embodiments, the analysis modules 308 may include modules configured to perform specific analyses with respect to a particular type of decision. Each module is executed by a processing system hosting an instance of the intelligence system 300. Non-limiting examples of analysis modules 308 include risk analysis modules, security analysis modules, decision tree analysis modules, ethical analysis modules, failure mode and effects analysis (FMEA) modules, hazard analysis modules, quality analysis modules, safety analysis modules, regulatory analysis modules, legal analysis modules, and / or other appropriate analysis modules.

[0215] In some embodiments, the analysis management module 306 is configured to determine which type of analysis to perform based on the type of decision requested by the intelligence service client 336. In some of these embodiments, the analysis management module 306 may include an index or other appropriate mechanism to identify a set of analysis modules 308 based on the type of decision requested. In these embodiments, the analysis management module 306 can receive a decision type and determine a set of analysis modules 308 to perform based on the decision type. Furthermore, or alternatively, one or more governance criteria may define when a particular analysis should be performed. For example, engineering criteria may define which scenarios require an FMEA analysis. In this example, the engineering criteria may be associated with a request for a particular type of decision, and the scenarios in which an FMEA analysis should be performed may be defined by the engineering criteria. In this example, the artificial intelligence module 304 performs a safety analysis module and / or a risk analysis module and determines an alternative decision if the action violates a legal or safety standard. After analyzing the proposed decision, the artificial intelligence module 304 selectively outputs the proposed conditions based on the results of the analysis performed. If the decision is permitted, the artificial intelligence module 304 outputs the decision to the requesting intelligence service client 336. If the proposed configuration is flagged by one or more analyses, the artificial intelligence module 304 determines an alternative decision and performs analyses on the alternative proposed decision until a suitable decision is obtained.

[0216] It should be noted here that, in some embodiments, one or more analysis modules 308 themselves may be defined by standards, and a combination of one or more related standards may constitute a particular analysis. For example, applicable safety standards may require a risk analysis in which multiple permissible methods can be used. In this example, to complete the risk analysis required by the safety governance standard, ISO standards for the entire process and documentation, and ASTM standards for narrowly defined procedures may be used in combination.

[0217] As described above, the aforementioned framework of the intelligence system 300 may be applied and / or utilized in various entities of the value chain. For example, in some embodiments, a platform-level intelligence system may be configured with all the functionality of the intelligence system 300, and specific configurations of the intelligence system 300 may be provisioned to each value chain entity. Furthermore, in some embodiments, if the intelligence service client 336 is unable to autonomously perform a task, it is configured to escalate the task to a higher value chain entity (e.g., edge level or platform level). In some embodiments, the intelligence service controller 302 may instruct lower-level components to perform intelligence tasks. Furthermore, in some embodiments, the intelligence system 300 may be configured to output a default action when a decision cannot be made by the intelligence system 300 and / or higher or lower-level intelligence systems. In some of these embodiments, the default decision may be defined in rules and / or standard libraries.

[0218] (Reinforcement learning for determining the optimal policy) Reinforcement learning (RL) is a machine learning technique in which an agent iteratively learns an optimal policy through interaction with the environment. In RL, the agent must discover the correct action through trial and error and maximize the concept of long-term reward. Specifically, a system employing RL consists of two entities: (1) the environment and (2) the agent. The agent is a computer program component connected to the environment that senses the state of the environment and takes action on it. At each step of the interaction, the agent senses the current state s of the environment and selects an action a to take. This action changes the state of the environment, and the value of this state transition is communicated to the agent by a reward signal r. The magnitude of r indicates the desirability (appropriateness) of the action. Over time, the agent constructs a policy π that defines the action the agent should take for each state of the environment.

[0219] Formally, in reinforcement learning, there exists a discrete set of environmental states S, a discrete set of agent actions A, and a set of scalar reward signals R. After learning, the system creates a policy π that defines the value of taking action aεA in state sεS. This policy defines the expected return value Qπ(s,a) when starting from state s, performing action a, and following policy π.

[0220] Reinforcement learning agents learn policies through repeated exposure to various states. The agent selects actions according to the policy and is rewarded based on a function that rewards the desired action. Based on reward feedback, the system "learns" the policy and is trained to generate the desired action. For example, in the case of a navigation policy, the RL agent repeatedly evaluates its own state (position, distance from the target object, etc.), selects an action (motor input for movement toward the target object, etc.), and evaluates the action using a reward signal that indicates the success of the action. (e.g., +10 reward if the distance between the moving system and the target object decreases due to movement, -10 reward if the distance increases due to movement). Similarly, in learning a grasping policy, the RL agent repeatedly acquires images of the target object to be grasped, attempts to grasp the object, evaluates the attempt, and uses the evaluation result of the previous attempt to decide on the next attempt to perform the next iteration.

[0221] Several approaches exist for policy training of RL agents. Imitation learning is a primary method, where the agent learns from state / action pairs that include actions chosen by experts (e.g., humans) in response to observed states. Imitation learning not only solves the problems of sample efficiency and computational feasibility but also makes the training process safer. RL agents derive multiple examples of state / action pairs by observing humans (e.g., moving to and grasping a target object) and use these as the basis for policy training. Behavior cloning (BC), which focuses on learning expert policies using supervised learning, is one example of an imitation learning approach.

[0222] The value-based learning approach aims to find a policy consisting of a set of actions that maximize the expected value of future rewards (or minimize the expected cost). Reinforcement learning agents learn a value / cost function and derive policies based on it. Generally, two types of expected values ​​are referenced: state value V(s) and action value Q(s,a). The state value function V(s) represents the agent's value in each state, and the action value function Q(s,a) represents the agent's value when performing action a in state s. The value-based learning approach works by approximating the optimal value (V* or Q*) and deriving the optimal policy from it. For example, the optimal value function Q*(s,a) can be identified by finding the sequence of actions that maximizes the state-action value function Q(s,a). The optimal policy for each state can be derived by identifying the highest-value action among the actions possible from each state. π*(s) = argmax Q*(s,a)

[0223] To iteratively calculate the value function as actions in a sequence are performed and the moving system transitions between states, the Bellman optimality equation can be applied. The optimal value function Q*(s,a) follows the Bellman optimality equation and can be expressed as follows: (eq.4) Q*(s t ,a t ) = E [r t+1 + γ max Q*(s (t+1) , a (t+1) )]

[0224] The policy-based learning approach directly optimizes the policy function π using an appropriate optimization method (e.g., stochastic gradient descent) and fine-tunes the parameter vector without calculating the value function. The policy-based learning approach is generally effective in high-dimensional or continuous action spaces.

[0225] Figure 7 illustrates a reinforcement learning-based approach that involves evaluating various states, actions, and rewards in determining the optimal policy for executing one or more tasks by a mobile system.

[0226] In 402, a reinforcement learning agent (e.g., an agent in the intelligence service system 300) receives sensor information, including multiple images taken by a mobile system in the environment. By analyzing one or more of these images, the agent can determine a first state related to the mobile system in 404. The data representing the first state includes environmental information such as images, sounds, temperature, and time, and mobile system information such as location, speed, and internal state (e.g., battery level, clock settings).

[0227] In 406, 408, and 410, various potential actions are determined depending on the state. Examples of potential actions include providing control commands to actuators, motors, wheels, flaps, and other components that control the agent's speed, acceleration, orientation, and position; changing the agent's internal settings, such as putting certain components into sleep mode to conserve battery life; changing direction when there is a risk of collision with an obstacle; acquiring or transmitting data; and attempting to grasp a target object.

[0228] Steps 412, 414, and 416 calculate the expected reward for each potential behavior based on the reward function. For each determined potential behavior, the expected reward is calculated based on the reward function. The reward may be based on a desirable outcome, such as obstacle avoidance, power saving, or data acquisition. If the behavior leads to a desirable outcome (e.g., obstacle avoidance), the reward will be high. Otherwise, the reward may be low.

[0229] The agent can also look to the future and analyze whether there are opportunities to achieve higher rewards in the future. In 418, 420, and 422, the agent determines the future state resulting from the potential actions in 406, 408, and 410, respectively.

[0230] For each future state predicted in 418, 420, and 422, one or more future actions are determined and evaluated. For example, in 424, 426, and 428, expected reward values ​​and other indicators associated with one or more future actions are derived. The expected rewards associated with one or more future actions are evaluated by comparing the values ​​of the reward function associated with each future action.

[0231] In 430, the behavior is selected based on a comparison of expected present rewards and future rewards.

[0232] In some embodiments, the reinforcement learning agent may be pre-trained through simulation within a digital twin system. In some embodiments, the reinforcement agent may be pre-trained using the behavior cloning method. In some embodiments, the reinforcement agent is trained using a deep reinforcement learning algorithm selected from Deep Q-Network (DQN), Double Deep Q-Network (DDQN), Deep Deterministic Policy Gradient (DDPG), Soft Actor Critic (SAC), Advantage Actor Critic (A2C), Asynchronous Advantage Actor Critic (A3C), Approximate Policy Optimization (PPO), and Trust Region Policy Optimization (TRPO).

[0233] In this embodiment, the reinforcement learning agent attempts to balance exploitation (utilizing existing knowledge) and exploration (investigating uncharted territory) when moving through the action space. For example, the agent follows an ε-greedy policy, occasionally randomly selecting exploration with probability ε, while taking the optimal action most of the time with probability 1-ε. Here, ε is a parameter satisfying 0 < ε < 1.

[0234] (Generative AI system) In exemplary embodiments, a generative artificial intelligence engine (GAIE) may be combined with a machine learning system in a transactional environment. Inputs to GAIE include images, videos, audio, text, program code, and data. Outputs from GAIE include structured and organized text, images, videos, audio content, software / programming source code, formatted data (e.g., arrays), algorithms, definitions, context-specific structures (e.g., smart contracts, transaction platform configuration datasets, etc.), and machine language-based data (e.g., API format content). In GAIE instances with models designed to process text data, GAIE can interact with other programmatic systems (such as conventional machine learning engines) to process other forms of data into text data. In exemplary embodiments, other programmatic systems, including a system running a machine learning algorithm, may generate text-based output (in large quantities as needed) that GAIE can consume. For example, consider another system that constructs thousands of text-based observations on other forms of data; this could be useful input for the GAIE model to learn from and process (e.g., summarize) into text-based output information. In an exemplary embodiment, the interface between GAIE and an integrated machine learning system may be extended to include system-to-system interaction. This interaction capability includes and / or access to the ability of GAIE to ask specific questions of the machine learning system and facilitate the refinement of its knowledge. For example, the interaction capability may include requesting the machine learning system to provide an assessment of its current market trading position. In another example, the interaction capability may convert numerical output from a machine learning engine into text (e.g., words such as high, medium, low) and provide it as input for interpretation by GAIE.

[0235] In exemplary embodiments, the data processed by GAIE may include one or more types of content. For example, GAIE may take as input one or more natural language representations, single or multidimensional shapes or models, real-world and / or virtual scene representations, LiDAR point cloud representations, sensor inputs and / or outputs, vehicle and / or machine telemetry, geographic maps, authentication credentials, financial transactions, smart contracts, resources such as processing instructions and / or shaders, device configurations such as HDL specifications for FPGA programming, databases and / or database structure definitions, or similar (including metadata associated with any data type). Inputs to GAIE may also include data representing one or more features of other machine learning models, such as configurations of other machine learning models (e.g., model type, parameters, hyperparameters), inputs, internal states (e.g., some or more weights and biases of a model), and / or outputs. These and other forms of content may be received as data in a variety of formats. For example, the natural language representations that GAIE receives as input may be encoded as one or more of the following: encoded text, images of written documents, recordings of human speech, videos of individuals using sign language, encoding based on embedded machine learning models, or similar, or any combination thereof. In exemplary embodiments, the inputs that GAIE receives and processes may include GAIE's internal states, such as partial results from partial processing of the input, or a set of GAIE's weights and / or biases as a result of prior processing (e.g., the internal states of a recurrent neural network (RNN)).

[0236] In some embodiments, the data and / or content received and processed by GAIE originates from one or more individuals, such as a person speaking natural language. In some embodiments, the data and / or content received and processed by GAIE originates from one or more natural sources, such as patterns formed by nature. In some embodiments, the data and / or content received and processed by GAIE originates from other devices (e.g., another machine learning model running on another device) or another component of the same device running GAIE (e.g., the output of another machine learning model running on the same device running GAIE, or a sensor in an Internet of Things (IoT) and / or cloud architecture). In some embodiments, the data and / or content received and processed by GAIE is artificially synthesized, such as synthetic data generated by an algorithm to augment a training dataset. In some embodiments, the data and / or content received and processed by GAIE is generated by the same GAIE, such as the internal state of GAIE for past processing and / or concurrent processing, or past output of GAIE in the manner of a recurrent neural network (RNN).

[0237] In some embodiments, at least a portion of the data and / or content received and processed by GAIE is also used to train GAIE. For example, a variational GAIE can be trained with inputs and corresponding acceptable outputs, and later receive the same inputs and output one or more variations of the acceptable outputs. In some embodiments, at least a portion of the data and / or content received and processed by GAIE is different from the data and / or content used to train GAIE. In some such embodiments, the data and / or content received and processed by GAIE is similar to, but different from, the data and / or content used to train GAIE, such as new inputs exhibiting a similar statistical feature distribution to the training data. In some such embodiments, the data and / or content received and processed by GAIE is different from and dissimilar to the data and / or content used to train GAIE. For example, new inputs with a significantly different statistical feature distribution from the training data. In scenarios involving dissimilar inputs, it is possible to determine the agreement between the first and second outputs by comparing one or more first outputs of GAIE for new inputs with one or more second outputs of GAIE for inputs in the training dataset. GAIE can request and / or receive additional training based on new inputs and their corresponding acceptable outputs. In scenarios involving dissimilar inputs, GAIE can provide alerts and / or explanations indicating how the new input and / or its corresponding output differs from previously received inputs and / or their corresponding outputs.

[0238] In the examples, GAIE's output may include one or more content types. For example, GAIE may generate as output one or more natural language representations, single or multidimensional shape models, real-world and / or virtual scene representations, LiDAR point cloud representations, sensor inputs and / or outputs, vehicle and / or machine telemetry, geographic maps, authentication credentials, financial transactions, smart contracts, resources such as processing instructions and / or shaders, device configurations such as HDL specifications for FPGA programming, databases and / or database structure definitions, or similar (including metadata associated with any of the data types). GAIE's output may also include data representing one or more features of other machine learning models, such as configurations of those models (e.g., model type, parameters, hyperparameters), inputs, internal states (e.g., some or more weights and biases of the model), and / or outputs. These and other forms of content may be generated by GAIE as data in various formats. For example, the natural language representations produced as output by GAIE may be encoded as one or more of the following: encoded text, character images, recordings of human speech, videos of individuals performing sign language, encodings based on embedded machine learning models, or similar, or any combination thereof. In exemplary embodiments, the output of GAIE may include the internal state of GAIE, such as partial results from partial processing of the input, or a set of GAIE weights and / or biases as a result of prior processing (e.g., the internal state of a recurrent neural network (RNN)).

[0239] In exemplary embodiments, a language-based dialogue-enabled GAIE may be configured to process various types of data to generate (e.g., describe) novel machine learning models that can provide novel and augmented text inputs for processing by the GAIE. In exemplary embodiments, a human can observe and interact with this ongoing dialogue between two systems. In exemplary embodiments, the dialogue is initiated by an expression from a conversational partner (e.g., a human or another device), and the GAIE generates one or more expressions in response to the conversational partner's expression. In exemplary embodiments, the GAIE generates an expression to initiate the dialogue and further responds to one or more expressions from the conversational partner in response to the initiation expression. In exemplary embodiments, the ongoing dialogue is turn-based, with each of the conversational partner and the GAIE generating an expression based on the other party's (conversational partner or GAIE's) previous expression. In exemplary embodiments, the ongoing dialogue is improvisational, with each of the conversational partner and the GAIE generating utterances regardless of the timing and / or order of preceding and / or simultaneous utterances of the conversational partner and / or the GAIE.

[0240] In exemplary embodiments, dialogue occurs between a GAIE and multiple conversation partners (e.g., two or more people, two or more other GAIEs, or a combination of one or more people and one or more other GAIEs). In some of these exemplary embodiments, the GAIE and each other conversation partner take turns generating expressions in response to preceding expressions from the GAIE and the other conversation partners. In some embodiments, one or more subconversations occur between the GAIE and a subset of the multiple conversation partners. Such subconversations may occur simultaneously (e.g., the GAIE has a first conversation with a first conversation partner and a second conversation with a second conversation partner at the same time) and / or sequentially (e.g., the GAIE has a first conversation with a first conversation partner and then a second conversation with a second conversation partner). Such subconversations may contain the same or similar topics or expressions (e.g., the GAIE presents the same or similar conversation-starting expressions to each of the multiple conversation partners and simultaneously has separate conversations with each of the multiple conversation partners about the same or similar topics). Such subconversations may contain different topics or expressions (e.g., GAIE presents different conversation starters to each of its multiple conversation partners and simultaneously engages in separate conversations with each of them on different topics). In the embodiment, the first conversation between the first subset of GAIE and its conversation partner may be associated with the second conversation between the second subset of GAIE and its conversation partner (for example, the second subset engages in the second conversation based on the content of the first conversation between the first subsets).

[0241] In exemplary embodiments, one or more of the GAIE and / or conversation partners may embody one or more roles. For example, the GAIE may generate expressions based on roles such as conversation initiator, conversation responder, teacher, student, supervisor, colleague, subordinate, team member, independent observer, researcher, specific character in a story, advisor, caregiver, therapist, supporter or facilitator of the conversation partner, or competitor or adversary of the conversation partner (e.g., a “devil’s advocate” who presents opposing and / or alternative views to the conversation partner’s beliefs or claims). In exemplary embodiments, at least one of the one or more conversation partners embodies one or more of the aforementioned roles or other rules. In exemplary embodiments, the GAIE’s role is relative to the roles of the conversation partners (e.g., the GAIE may embody the roles of boss, colleague, or subordinate relative to the roles of the conversation partners). In exemplary embodiments, the GAIE’s role in a first conversation between a first subset of the GAIE and multiple conversation partners may be identical or similar to the GAIE’s role in a second conversation between a first subset of the GAIE and multiple conversation partners. In exemplary embodiments, the role of GAIE in a first conversation between a first subset of GAIE and multiple conversation partners may differ from the role of GAIE in a second conversation between the first subset of GAIE and multiple conversation partners (e.g., GAIE embodies the role of a teacher in the first conversation and the role of a student in the second conversation). In exemplary embodiments, GAIE's role in a conversation may change over time (e.g., GAIE may initially embody the role of a student in a conversation and then change to the role of a teacher within the same conversation). In exemplary embodiments, GAIE may embody two or more roles in a conversation (e.g., GAIE may represent two personalities in a conversation, each representing two characters in a story). In exemplary embodiments, GAIE may generate representations between two or more roles in a conversation (e.g., GAIE may generate a dialogue between two characters in a story). In exemplary embodiments, GAIE may participate in multiple conversations, each using the same or similar modality (e.g., participating in multiple text-based conversations simultaneously).In an exemplary embodiment, GAIE can participate in multiple conversations using different modalities (e.g., participating in the first conversation via text and the second conversation via voice).

[0242] In exemplary embodiments, a GAIE participating in a conversation is associated with an avatar (e.g., name, color, image, two-dimensional or three-dimensional model, voice, etc.). Expressions generated by the GAIE are presented as if they originated from the GAIE (e.g., in the voice associated with the GAIE, or in a speech bubble displayed near the GAIE's visual location in a virtual or augmented reality environment). In exemplary embodiments, the GAIE's avatar may be configured based on the GAIE's role (e.g., a GAIE embodying the role of a teacher is associated with an avatar depicting a teacher). In exemplary embodiments, the GAIE's avatar may be incorporated into a real-world actor (e.g., a robot in a real-world environment such as a stage performance).

[0243] In an exemplary embodiment, GAIE may include a generative pre-trained transformer element that can be configured as a language model designed to understand various types of input and generate chat commands for a chat-type interface system. These commands may include software development tasks, API calls, and the like. In an exemplary embodiment, such a language model may include input functions that support the reception of images (including videos) to construct text outputs, functions, and additional questions that can be inserted into the dialogue between the two systems in the above dialogue embodiment. In an exemplary embodiment, this multimodal support enables contextual analysis of images and other media formats. For example, a user / customer can upload images and other media to a GAIE-enabled platform. Based on aspects of the corresponding input prompts, the multimodal GAIE can be configured for use in an evaluation workflow to identify macro and micro attributes from multiple perspectives and their correlational effects on the evaluation. In this example, a photograph / image of an old car is input along with evaluation-related prompts. In response, GAIE may identify one or more typical values ​​based on the detected car attributes such as manufacturer / model. GAIE can further consider finer details within the image to suggest indicators that may be factors influencing the value. For example, fine details in images, such as damaged body panels, could potentially lower the vehicle's value below average. In another example, fine details in images showing markings that match a limited production lot could potentially increase its valuation.

[0244] In the examples, the GAIE in the subject area can be adapted to facilitate transaction forensics. As AI-driven transactions increase, the need for humans to understand how and why specific transactions were initiated and executed is expected to grow. For example, a transaction may be generated in response to a user request such as "Please send me a new circuit board for a broken refrigerator." If the requested circuit board arrives with, for example, a tracking device from a hostile government embedded in it, it may be beneficial to reveal how the AI ​​system executed the transaction that procured that circuit board. It would also be beneficial for the AI ​​system to participate in establishing AI system control actions and procedures that can be taken to prevent the recurrence of unacceptable procurement practices.

[0245] In transactions involving collateral or insurance, GAIE can be configured to provide support such as collateral valuation and defining and fulfilling insurance needs.

[0246] A GAIE that has pre-trained on trading targets may present the following information in response to token acquisition-related prompts from investors: an explicitly stated set of goals, a set of opportunities for new token acquisition candidates, a set of comparative advantages over other tokens, and the potential relevance of the token's strengths to the investor's goals. In an embodiment, a system with a portfolio analysis engine is combined with a conversational engine that discovers investment opportunities based on the user's investment goals and generates a summary of those opportunities to present to the user. This summary includes why the investment opportunity aligns with the user's investment goals. In various embodiments, the summary is based on one or more of the user's characteristics (e.g., the user's financial situation, the user's demographic characteristics, the user's understanding of trading, portfolios, markets, and the economy, and / or the user's past trading history related to portfolios, markets, and the economy).

[0247] Adaptive GAIEs can facilitate the generation of synthetic transaction data from disposable training models that can be discarded after training. Synthetic data from the original sources incorporated into the trained GAIE can be regenerated with personal information removed, overcoming privacy concerns and facilitating data sharing and / or pooling among transaction entities (e.g., banks and third parties). In an exemplary embodiment, the scope of application for GAIE includes interaction with a transaction engine that processes transactions between multiple entities using synthetic data generated by GAIE (derived from a training set of historical transaction data). In an exemplary embodiment, the data used to train GAIE is stored for future use. For example, the training data can be subsequently validated and used to identify the causes of GAIE's output and behavior. For example, if GAIE exhibits bias or defects, characteristics within the training data can be investigated to identify the causes of the bias or defects in GAIE. Subsequently, additional training data can be provided to perform continued training or retraining of GAIE, which complements the characteristics that caused the bias or defects in GAIE.

[0248] In exemplary embodiments, a GAIE fine-tuned to a transaction subject could offer a wealth of functional improvements, such as transaction subject-related searches and digital wallet searches. In exemplary embodiments, a generative AI conversational agent could be configured to search a set of digital wallets.

[0249] In exemplary embodiments, a GAIE may be pre-trained to perform financial system management functions, such as "smart treasury management," within an Enterprise Access Layer (EAL) system. For example, an EAL pre-trained GAIE can describe, predict, and / or determine expected yield generation across different accounts, independently of on-chain and off-chain interactions that affect yields. A pre-trained smart financial management GAIE sets risk tolerance and target parameters and interacts with the learning system through pre-training on a transaction (e.g., financial) data pool. In examples, such a pre-trained GAIE is not limited to financial management but can be applied to the management of any asset aiming to generate returns with a cross-system parameter set. In exemplary embodiments, such a GAIE may include, or interact with, presentation layer functionality (e.g., a data story engine) to concisely provide asset management information to users across accounts. In exemplary embodiments, such a GAIE may generate content, such as data stories, based on simulation information about different event-based outcomes aggregated across multiple accounts.

[0250] In an exemplary embodiment, an EAL-pre-trained GAIE can be trained to create, configure, or manage enterprise data pools used across the enterprise's (or on behalf of the enterprise's) transactional systems. Other capabilities of an EAL-pre-trained GAIE include workflow development, transactional workflow configuration, utilization, reuse, and creation of workflows and tasks, fraud analysis, employee training at various levels, including expert level, and transactional complexity reduction.

[0251] In exemplary embodiments, such a GAIE may facilitate process workflow orchestration, using a conversational generative AI agent and another AI-assisted process in a coordinated sequence. In exemplary embodiments, a GAIE may generate, execute, maintain, and / or supervise one or more workflows in a robotic process automation (RPA) environment. For example, a GAIE may be trained to monitor an individual's facial expressions and behaviors during interactions with others, and then generate similar facial expressions or perform similar behaviors in similar interactions between the GAIE and others. In some such scenarios, the GAIE passively observes the individual during interactions with others and self-learns to behave similarly to that individual in similar interactions with others. There are also scenarios where the individual actively trains and instructs the GAIE (e.g., creating and performing exemplary / educational dialogues for the GAIE), and based on that training and instruction, the GAIE takes similar behaviors in subsequent interactions with the individual. In exemplary embodiments, a GAIE may be trained and / or instructed by an individual to perform a specific behavior during an interaction with that individual, and then perform that behavior during an interaction with the same individual who provided the training and / or instruction.

[0252] In an exemplary embodiment, the enterprise access layer (enterprise access tier or corporate access tier) has an intelligent agent that learns workflows performed by a group of users in a semi-supervised learning manner based on user interactions, and the intelligent agent performs at least one step in the learned workflow. In an exemplary embodiment, the intelligent agent automatically requests feedback from one or more users to complete the workflow steps and improve its training.

[0253] The application areas of the EAL-pre-trained GAIE platform include data pooling, intelligence system management, workflow development, expert training, fraud analysis, requirements refinement, and governance. Examples of these areas are shown below.

[0254] In the data pool application domain, EAL pre-trained GAIE configures, curates, builds, and manages access to static and mobile data pools that facilitate use cases, customers, agents, and other EAL workflow requirements. In the intelligence system management domain, GAIE enhances systems with supervised generative AI capabilities that determine how and when to apply various AI tools / modules. In the workflow development application domain, pre-trained GAIE identifies, refines, and generates various transactional workflows, such as data and finance, enabling modularization, reuse, and further data-driven improvements. In the expert training application domain, GAIE collaborates with experts and approvers to build domain-specific capabilities that can be used for workflow enhancement, governance, and fraud detection. In the fraud analysis application domain, GAIE collaborates with fraud experts, criminal records, and past fraud convicts to enhance detection capabilities. In the requirements refinement application domain, GAIE refines all requirements and transactions, reducing computing and data transmission resources. In the governance application domain, pre-trained GAIE helps determine "when," "where," and "what" is relevant to governance requirements.

[0255] In exemplary embodiments, a GAIE may be pre-trained for customer verification / transaction verification (KYC / KYT) applications. In exemplary embodiments, such a pre-trained GAIE may generate customer profile summaries based on contextual analysis of information obtained from social media, etc. Such a pre-trained GAIE facilitates iterative tracking / observation of conversations and user behavior to determine the impact of conversation parameters on user behavior (group / cohort level). It also facilitates iterative tracking / observation of conversations and user behavior to determine the impact of conversation parameters on user behavior (e.g., at the individual level).

[0256] From the perspective of the trading environment and / or related smart contracts, a pre-trained GAIE can assist in constructing smart contract clauses based on interactive dialogue with customers. Such a pre-trained GAIE can also generate and, if necessary, negotiate intellectual property license clauses. In an exemplary embodiment, a smart contract generation system may include a GAIE-based system configured to take in and interpret contract-related clauses (e.g., those instructed by an individual) and generate corresponding smart contract configuration data structures, wrappers, etc. A system for flagging non-standard smart contract clauses / conditions may include a generating AI conversational agent configured to process contract clauses and flag non-standard aspects of smart contract clauses and / or conditions. In an exemplary embodiment, a system based on a pre-trained GAIE can develop a set of scope definitions for smart contracts and / or connect scope definitions to proprietary standards and data.

[0257] As an example, a pre-trained GAIE might involve the intelligent recursive use of an AI assistant based on the results of an initial query (e.g., a prompt) that may require access to proprietary or purchased standards or data. Such an AI assistant could embody one or more roles, such as a personal data assistant (PDA), teacher, student, supervisor, colleague, subordinate, team member, coach, independent observer, researcher, specific character in a story, advisor, caregiver, therapist, supporter or facilitator of a conversation partner, or even a competitor or adversary of a conversation partner. In this example, the GAIE might receive a prompt requesting a smart contract scope that includes chemical compatibility testing of a group of plastics used in flow batteries. The initial query is adapted and / or regenerated (e.g., from a pre-trained GAIE) as a prompt to identify the appropriate plastic chemical compatibility testing standards that require access. Depending on the access rights obtained, the GAIE can develop a modified scope based on the regenerated query and write a smart contract to perform the tests based on the modified scope.

[0258] In an embodiment, the pre-trained GAIE system includes a smart contract analysis engine that determines one or more features of a smart contract under consideration by a user. Further, the GAIE includes a conversation engine that explains those features to the user, such as summarizing the content of the smart contract.

[0259] In an exemplary embodiment, the GAIE may be pre-trained to perform prompt generation based on a data story or a plurality of cross-system information sources. Exemplary generation prompts include instructing and / or requesting the pre-trained GAIE to tell a story about a "journey" regarding products, business relationships, events, service providers, smart container fleets, robot fleets, and the like.

[0260] In an exemplary embodiment, GAIE can receive a story plot or ending and generate content that follows that plot or produces that ending. In an exemplary embodiment, GAIE can generate a story plot or ending and further generate content that is consistent with the story plot or ending generated by GAIE. In an exemplary embodiment, GAIE can receive a story world or environment and generate content that takes place within a given world or environment. In an exemplary embodiment, GAIE can generate a story world or environment and further generate content that takes place within a world or environment generated by GAIE. In an exemplary embodiment, GAIE can receive a character or event to be included in a story and generate content that includes a given character or event in the story. In an exemplary embodiment, GAIE can generate a world, environment, character, event, etc., "from scratch" (for example, based on randomized input). In an exemplary embodiment, GAIE can generate a world, environment, characters, events, etc. (for example, a story based on real public figures or events) based on a given world, environment, characters, events, etc.

[0261] In exemplary embodiments, GAIE can receive a first story and generate a second story related to the first story. For example, GAIE can generate a second story that is an alternative retelling of the first story (e.g., a second story that retellings the first story from the perspective of a different character than the narrator of the first story). GAIE can generate a second story that takes place in the same or similar world or environment as the first story, or a second story that takes place in a different world or environment related to the world or environment of the first story. GAIE can generate a second story that features characters and events from the first story, or a second story that features different characters and events related to the characters and events of the first story.

[0262] In an exemplary embodiment, GAIE can generate a story from the perspective of the narrator or an independent observer (e.g., a third-person narrative). In an exemplary embodiment, GAIE can generate a story from the perspective of a character or point of view within the story (e.g., a first-person narrative), including characters generated and / or embodied by GAIE. In an exemplary embodiment, GAIE can generate a story from the perspective of a listener or audience from whom the story is presented (e.g., a second-person narrative). In an exemplary embodiment, GAIE can generate a story from multiple perspectives, for example, Part 1 of the story being generated from the perspective of a first character, Part 2 from the perspective of a second character, and Part 3 from the narrator's perspective. In an exemplary embodiment, GAIE can generate a story containing a sequence of two or more events (e.g., a story containing two or more events observed by a character). In an exemplary embodiment, GAIE can generate a story containing events described from multiple perspectives (e.g., a story describing an event from the perspective of a first character and the same event from the perspective of a second character).

[0263] In an exemplary embodiment, GAIE may generate a static story that remains unchanged even when retold. In an exemplary embodiment, GAIE may generate a dynamic story that changes when retold (e.g., adding details with each retell). In an exemplary embodiment, GAIE may modify a story based on user input (e.g., outcome selection by the story's recipient). In an exemplary embodiment, GAIE may generate a story based on one or more inputs received from one or more recipients of the story (e.g., based on prompts based on user requests (e.g., a request to create a story that includes a specific event specified by the user)). In an exemplary embodiment, GAIE may receive feedback from recipients about a story (e.g., expressions of pleasure, displeasure, agreement, disagreement, satisfaction, dissatisfaction, confusion, etc., regarding the characters, events, or characteristics of the story), and GAIE may update the story based on that feedback (e.g., adding, removing, or clarifying events in the story, or switching the perspective (point of view) of an event from a first character in the story to a second character).

[0264] In exemplary embodiments, GAIE is trained by loading data (such as structured and unstructured data where numerical or non-text values ​​may be dominant). Examples of such training data include one or more database schemas. Techniques for curating and integrating purpose-specific data (including curating models as input to GAIE) include curating domain-specific data, data, and model discovery.

[0265] Potential areas of innovation enabled by or related to advancements in GAIE include user behavior models (optionally including feedback and personalization), group clustering and similarity, personality type classification, input and process governance, justification of GAIE knowledge and proof points, genetic programming with feedback capabilities, intelligent agents, voice assistants and other user experiences, transaction agents (finding trading partners and negotiating), agents interacting with other agents, opportunity miners, automated discovery of agent generation and application opportunities, user interfaces that adapt to users and context, hybrid content generation, collaborative units of humans and generative AI, purpose-specific data integration, selected data sources, and curation of data modeled as input to generative AI.

[0266] In embodiments of GAIE-enabled systems (e.g., for robotic process automation), the GAIE system can summarize a series of actions to be automated and explain the context of those actions. For example, it might say, "Based on the following features, we have determined that these properties meet your requirements. Which property is the most attractive?" In this way, the GAIE-enabled process automation system can request feedback for rapid learning based on feedback.

[0267] In exemplary embodiments, emerging features of GAIE technology can significantly improve upon previous versions, for example, by integrating domain-specific knowledge (such as mathematics) with the chat interface. Furthermore, emerging features include improved understanding and processing capabilities for prompting on complex topics. In addition, knowledge organization is also significantly improving as GAIE systems evolve. In exemplary embodiments, an updated GAIE can correctly respond to a prompt asking for "today's date," whereas a previous version might respond with "today's date (e.g., current date) may be the date GAIE was last trained."

[0268] In exemplary embodiments, a pre-trained (e.g., subject-specific) GAIE can provide better personalization than a base GAIE instance. Generally, a base GAIE may attempt to personalize responses when explicitly informed of user details, while a subject-specific, or other pre-trained GAIE, can be configured to provide a unique latent context for interactions that include user-personalized responses by incorporating and / or making accessible structured information about the user (e.g., determined based on user identification and / or prompt-based cues) as a component.

[0269] In exemplary embodiments, the GAIE is configured to support the interpretability and / or explainability of its output. In exemplary embodiments, along with the output, the GAIE provides a description of the basis of the output, such as an explanation of why it produced this particular output as a response to the input. In exemplary embodiments, along with the output, the GAIE provides a description of the internal state of the GAIE that led to the output. This includes the set of variational parameters of the variational encoder that were processed in combination with the input to produce the output, and / or the internal state of the GAIE due to the GAIE's prior processing that led to the output (e.g., similar to a recurrent neural network (RNN)). In exemplary embodiments, along with the output, the GAIE provides an indication of one or more subsets of the input features that are particularly associated with the output (e.g., in a GAIE that outputs an image caption or summary, the GAIE may also identify specific parts or elements of the image that are related to a part of the caption or summary).

[0270] In exemplary embodiments, an advanced GAIE, such as one pre-trained for subject-specific operations, may be trained for epistemological improvements to help determine evidence for content presented as facts in the provided response. One example of epistemological improvement is citing knowledge sources related to facts in the response as a means of proving those facts. This is essentially a way in which the GAIE “shows the process,” or at least the origin of that process. In exemplary embodiments, the GAIE generates an output based on information received from one or more external sources (e.g., messages in a message set, websites on the internet) and indicates some of the information related to that output (e.g., websites on the internet that provided the information included in the GAIE’s output).

[0271] The advanced GAIE described and conceived herein can maintain context awareness throughout the entire chat (user prompt / GAIE response) interaction. Maintaining context awareness prevents GAIE from starting each chat session from scratch without context regarding past chats with the same user. It also allows GAIE to continue resuming conversations from previous interactions between the user and the GAIE. Furthermore, maintaining context awareness and elapsed time awareness between dialogue sessions allows responses to prompts in resumed chat sessions after interruptions to be adapted to elapsed time and situational changes based on trained knowledge. For example, GAIE can determine that a deadline mentioned in a previous chat has expired, or that a significant intervention event has occurred (such as a local team losing a crucial match). Moreover, context awareness between temporally separated chat sessions is extremely useful in projects where real-world physical constraints exist (e.g., when human evaluation, discussion, and decision-making in smart contract negotiations involve time-dependent factors such as involvement in other priorities). This can determine whether each conversation is treated as separate / partitioned / isolated, or whether continuous, temporally separated conversations are treated as resumable as if (almost) no time had passed. In an exemplary embodiment, GAIE may be configured with a contextualization module that holds the concepts of referential conversation sessions and interconnections for detail and continuity (e.g., yesterday's conversation). This contextualization allows for the avoidance of repetitive responses and improves the efficiency of referencing past conversations. Furthermore, the contextualization module may provide GAIE with the context of other conversations, such as between a user and a system, or between other users and systems.

[0272] In instances where such context is maintained, a context-aware GAIE can provide weather forecast responses that refer to an earlier time. For example, a context-aware GAIE could provide a weather-related response such as: "We discussed the weather on Monday, and on Wednesday, I was asked if I would need an umbrella. Based on the forecast at the time, I answered 'probably not.' According to the updated weather forecast, the probability of rain on Wednesday has increased, so you will probably need an umbrella."

[0273] Other features of the emerging GAIE system may include the adaptation of GAIE to the generation and manipulation of digital avatars. In exemplary embodiments, digital avatars may be programmed with their own unique visual representations. To improve the similarity between avatars and their owners based on the user's appearance and voice interpretation, the GAIE training / pre-training dataset may require information on body language and nonverbal cues, such as gaze, posture, and voice pitch / volume.

[0274] Emerging GAIE systems may include, for example, the determination of responses based on user activity and the adaptation of variations and nuances. A user's physical state (sitting, walking, driving, exercising, etc.) may influence GAIE content generation (e.g., presentation of different signals). Furthermore, GAIE systems can adapt responses to prompts in different environments, such as real-world dialogue, voice interfaces, and virtual reality, according to their variations and nuances. Other aspects that may influence GAIE responses to prompts include differences and nuances such as different cultures and demographics. In addition, in exemplary embodiments, advanced GAIE training and operational methods / systems may include the recognition of the user's advanced communication characteristics (e.g., humor, sarcasm, insincerity, double meanings) and emotional states.

[0275] In exemplary embodiments, methods and systems for enhancing a GAIE platform as described herein may include configuring GAIE participation in multi-user conversations. In group environments, strict turn-based conversations with one person are difficult, and the context of who is speaking to whom in each expression becomes important. The contrast between more fluid multi-user conversation structures and turn-based structures suggests that advancements in GAIE may include the development of an understanding of: social interaction and cues (e.g., who each expression is directed to), group dynamics (e.g., who is the group leader?), interpersonal relationships, the concept of branching threaded arguments, concurrent discussions between various subgroups within a group, when to interject without interrupting other users, a sense of conversational balance to avoid dominating the conversation, consideration, user sensitivity to personal information, and an understanding of what can and cannot be shared in a group setting, based on context and relationships with other users.

[0276] Regardless of whether the interaction is one-to-one or multi-user, GAIE is expected to evolve beyond the turn-based paradigm. For example, current GAIEs can generate media (images, music, videos, etc.) based on user prompts (which may themselves be one or more media types) and refine the generated media based on user interactions (such as modifying content in specific ways or extending the boundaries of an image with additional content consistent with existing content (e.g., outpainting)). More advanced generative AIs can flexibly and continuously adapt generated content to contextual user input and interactions. For example, GAIEs can adapt media generation in response to user integration with generated media content, such as allowing the user to virtually walk around within the content and interact with and react to content items. Such media-adaptive GAIEs can generate new content or update content based on user input / virtual content interaction. Furthermore, to facilitate immersive virtual interaction with generated content, detailed information about the user can be considered as part of the criteria for generating and / or updating media.

[0277] In this embodiment, a GAIE with user-immersive interaction and media output without feedback generates media (e.g., Image 1) based on a prompt from the user specifying the theme of the story. The user can then specify a series of subsequent scenes, and the GAIE generates images corresponding to each scene, forming a series of storyboards for the narrative.

[0278] When the media output - compatible GAIE collaborates with the user's immersive function, the user can, for example, operate an avatar, move within the scene, and interact with the generated media objects. For example, based on the order and method in which the user moves within the scene and interacts with the objects, the generation algorithm generates new content (e.g., after the user views a specific painting on the wall of a gallery, the user opens the window curtains). Outside the window, there may be an entire world that is consistent with the specific painting the user viewed. If the user chooses to move the avatar into that world, the painting on the wall is updated to reflect the user's interaction.

[0279] As another example of immersive user - generated media content, the user may request a science - fiction story. In addition to generating a story based on tropes commonly associated with science - fiction, GAIE may include tropes familiar to the user, such as an SF version of a character well - known in the mythology or literary works to which the user belongs, based on the user's age, culture, and other interests. In some cases, the algorithm may even include in the created story people similar to famous or public figures in the user's culture or generation, or the user's own friends and acquaintances.

[0280] As a specific example, GAIE may be pre - trained for market construction (such as setting up a new marketplace, discovering trading partners, ecosystem - based trading, aggregating demand and supply, negotiating contract terms, setting up smart contracts, mediating transactions, generating simulations for the digital twin of an exchange, personalizing financial / trading advice, etc.).

[0281] As an example of GAIE adapted to market adjustment responses, a generative AI dialogue agent enables the setting up of a new marketplace. In another example of GAIE adapted to market adjustment responses, a generative AI dialogue agent is configured for discovering trading partners, assets, and / or marketplaces.

[0282] As an example of GAIE adapted to market orchestration, the generative AI interactive agent can be configured for ecosystem-based trade presentations. As an example of GAIE adapted to market orchestration, the generative AI interactive agent can be configured for demand and / or supply aggregation. As an example of GAIE adapted to market orchestration, GAIE can be configured for contract term negotiation. As an example of GAIE adapted to market adjustment, GAIE enables the configuration of smart contracts. As an example of GAIE adapted to market adjustment, generative AI interactive agents can be configured to mediate trades. As an example of GAIE adapted to market adjustment, the generative AI conversational agent can be configured to generate simulations for the exchange's digital twin. As an example of GAIE adapted to market adjustment, the generative AI conversational agent can be configured to generate personalized financial and / or trading advice.

[0283] As an example of a GAIE adapted to a game environment, a generative AI interactive agent can be configured to generate a game environment and / or experience (e.g., by using a game engine). For example, a GAIE adapted to a game environment may be configured to generate a personalized game environment and / or experience. For example, a GAIE adapted to a game environment may generate NPC text / conversation. This allows a game environment with a non-player character text generator to interactively communicate relevant game objective-driven data to the human player of the game using AI / machine learning. In an exemplary embodiment, a GAIE adapted to a game environment may include a game engine and a conversational agent that navigates the customer journey using contextual, generative conversational AI based on conversational comparison with the customer journey script. In an embodiment, the GAIE may be integrated with the game engine.

[0284] In exemplary embodiments, the superintelligence system may be based on a pre-trained GAIE that facilitates the automated discovery of relevant domain-specific knowledge and examples. The superintelligence system may further leverage the pre-trained advanced GAIE to generate content using domain-specific examples. Furthermore, the superintelligence system may include genetic programming capabilities for generating novel mutations. In exemplary embodiments, the superintelligence system may further include a feedback system (e.g., collaborative filtering and automated outcome tracking) to narrow the mutations to favorable outcomes (e.g., financial, personalization, group targeting).

[0285] In an exemplary embodiment, GAIE may be pre-trained for use and / or collaborative operation with a digital twin engine, including instances such as a digital twin for senior executives. In an exemplary deployment, GAIE interacts with the digital twin and provides a narrative about the digital twin's topic to the audience. In this example, the digital twin interacts with GAIE (e.g., through an API) to generate a summary narrative for the CEO and a detailed narrative for the CFO.

[0286] Executive digital twins are configured for specific roles or users. Therefore, a GAIE system with a digital twin interface can enhance the functionality of executive digital twins by curating the data consumed by executive digital twins with different roles and filling them with content. For example, GAIE might receive information about the executive digital twin and the person it represents (e.g., the user's role). GAIE determines the narrative level of detail for each executive digital twin. This is based on a general digital twin / user role criterion or is tailored through interaction with the specific user of that digital twin. In an exemplary embodiment, a technology-oriented CEO might receive a more "detailed" narrative about technology and research and development (R&D), while a CEO with a finance background might receive a narrative with less detail about technology-related functions, focusing more on financial analysis.

[0287] In an exemplary embodiment, a GAIE system interacting with a digital twin engine (e.g., a digital twin instance and / or engine for executives) can distinguish between relevant and irrelevant or noise-based content from the entire pool of potential content being learned, for a specific narrative topic, target human consumer, etc. Based on this relevance determination, the GAIE system generates output data according to the relevant data and the determined level of detail.

[0288] Furthermore, the GAIE system can select real-time data sources and connect them to the target / requesting executive digital twin. GAIE then configures consumption pipelines on the fly for those sources (e.g., identifying data sources, requesting data from identified data sources, configuring APIs, etc.). Therefore, in this example, the GAIE system will identify the data sources and connect them to the executive digital twin instances / engines.

[0289] One example of a use case is a digital twin for executives that provides access to complete financial data for historical periods (e.g., previous year / quarter / month). The executive digital twin enables GAIE to access all of this data. GAIE then determines the level of detail of the data for its target audience (e.g., target consumers of narratives on topics included in the complete financial data).

[0290] If the target consumer / viewer holds a CEO role, GAIE may decide to include key insights but omit detailed information in the CEO-focused narrative. GAIE will then generate a narrative of key insights for the target period (e.g., the current quarter) based on at least the received data.

[0291] A pre-trained GAIE can be used for describing the attributes of a digital twin, describing interactions with other digital twins and the environment, describing simulations, generating content using digital twin simulation data, realizing context-adaptive digital twins for managers, and facilitating the development of narratives about ongoing real-time operations (adjusting to the preferred conversational style of the user represented in the digital twin), enabling the creation, management, and manipulation of digital twins. In an exemplary embodiment, a context-adaptive running digital twin integrated with a generative conversational AI system may be configured to generate a series of narratives about the operations of a company based on an input dataset of real-time sensor data obtained from the company's operations. The digital twin (or human user) can prompt the GAIE and / or the conversational AI system to compare financial data with real-time sensor data.

[0292] GAIE can be adapted (e.g., pre-trained) to facilitate the enhancement of AI training data related to digital twin applications. In an exemplary embodiment, the method may include the creation of synthetic training data using an AI conversational agent.

[0293] Furthermore, in relation to digital twin technology, GAIE can be adapted to summarize high-granularity data in a form that is available to executive digital twins. In this regard, the executive digital twin system may include an intelligent agent that receives a set of customization features from the user (e.g., executives represented in the digital twin), including the user's role within the organization. The intelligent agent can also determine the level of granularity of the report based on the customization features. In an exemplary embodiment, the set of customization features includes granularity specifications for different types of reports. Furthermore, the intelligent agent determines the level of granularity of the report based on the user's role within the organization. The subject matter of the report may also be generated based on the user's role within the organization.

[0294] In an exemplary embodiment, a voice-based user interface for customizing the level of specificity of executive digital twin report generation can be operatively coupled to a customized GAIE. This GAIE processes voice into a series of report instructions (and optionally report content) based on the attributes of the user(s). Examples of voice-based requests processed as described above include "I want an executive summary level report on predictive maintenance" and "I want a detailed report on competitive analysis". The voice-based user interface responds to such requests and instructs the corresponding executive digital twin system to supply the generative AI engine (e.g., GAIE) with the level of parameter specificity as additional input in addition to the data. In this example, IoT data from a manufacturing facility is utilized for predictive maintenance. The response to prompts regarding preventive maintenance can customize the level of detail based on the role of the target report consumer, such as an operations-based role. The level of detail includes cost, required maintenance time, predicted downtime, methods and timings for offsetting maintenance activities, etc. For a finance-based role, it is possible to adapt specific levels, such as the impact on short-term revenue, impact on the supply network, impact on market share, impact on stock price, etc.

[0295] When a digital twin models an individual, a finely tuned GAIE (Artificial General Intelligence) can be used to coordinate the collaboration between the digital twin and the human, improving accuracy (e.g., if human behavior or reactions differ from the digital twin's predictions, the GAIE can initiate a dialogue with the user to identify the cause and use the results to update the individual's digital twin model). Instead of human experts occasionally participating in automated digital twin model training (e.g., to correct errors or provide new examples), the corresponding GAIE can periodically query the user to gather additional information for updating the individual's digital twin model. For example, the system could include a digital twin modeling an individual and further include a conversational engine that facilitates decisions on updating the digital twin based on the individual's dialogue with the person regarding the differences between the individual's and the digital twin's predicted behaviors.

[0296] In this example, the GAIE system is configured for use in an automated manufacturing environment. For instance, the user prepares a prompt describing the product they wish to 3D print. The GAIE system generates a series of instructions for 3D printing, including machine settings for the automated 3D printer and renderings showing the resulting printer output. In another example, the user could include a photo of the product as a prompt, requesting a modified 3D print instruction such as, "I want this bicycle, but I want to change the tires and make it red."

[0297] Another use case for a pre-trained GAIE is generating guidance recommendations for energy saving, usage shifts, and other related topics using user behavior data. In particular, recommendation systems for energy saving, usage shifts, and optimization can include integrated generative conversational AI systems that adapt their output based on user behavior from user behavior datasets.

[0298] As a concrete example, adaptive GAIE can support energy resource management. Energy resource management systems can be enhanced to provide advanced intelligence (e.g., superintelligence) for planning, managing, and controlling distributed energy resources (DERs) and energy generation, storage, consumption, and transmission facilities. Components of a superintelligence energy management system include automated discovery of relevant domain-specific knowledge and examples, generative AI that leverages domain-specific examples to generate content, genetic programming that generates novel mutations, and feedback systems (e.g., collaborative filtering, automated result tracking) that narrow mutations towards favorable outcomes (financial, personalization, group targeting, etc.). As an example, a superintelligence AI-powered management system can be configured to manage multiple systems on an energy edge platform through automated discovery, generative AI, genetic programming, and feedback systems.

[0299] In an exemplary embodiment, the GAIE is adapted to the patent field (e.g., through training and pre-learning) and generates patent claims in response to the provision of patent disclosures. The activated GAIE receives patent claims as prompts and generates supporting patent disclosures therefrom. In an exemplary embodiment, the activated GAIE is trained to understand patent and claim structures in multiple jurisdictions.

[0300] In an exemplary embodiment, the GAIE may be pre-trained (e.g., fine-tuned) using private instances of a company's intellectual property data (e.g., products, business objectives, competitive considerations, core inventive ideas, etc.). In an exemplary embodiment, private instances of company data for patent generation may be configured (e.g., as prompt-response pairs) for fine-tuning the GAIE instance.

[0301] In addition to patent disclosure and drawing creation, GAIE can be fine-tuned to generate drawings, draft disclosure statements from drawings, draft claims from drawings, respond to examiner notices, collect evidence of use (EOUs) for patent monetization, create a patent claim matrix for the entire portfolio, generate high-level landscape search strings, and enhance search strings. Fine-tuning may include preparing a set of prompt responses for various intellectual property-related actions, such as asserting patent claims, infringement analysis and evidence disclosure, claim acceptance and / or rejection, claim scope estimation, and claim quality. In exemplary embodiments, an intellectual property-specific GAIE may be pre-trained with information from proceedings related to infringement litigation to understand potential infringement, etc.

[0302] GAIE's training and IP integration facilitate the translating broadly described inventive concepts into disclosures that reflect robust implementability and / or support. For example, a summary can serve as an input prompt for creating patent application documents (disclosures, drawings, abstract, summary, and optional claims). The generated results may become part of subsequent prompts, along with a description of the innovation's general theme, category, focus area, and / or other classifications. For example, describing a trading environment processing platform might require examples of technical implementations, systems, and / or method designs. For instance, "In the context of the aforementioned trading environment processing platform, what types of hardware and software could be used to implement the trading environment governance engine?"

[0303] Regarding development processes focused on monetizing intellectual property (patents, etc.), GAIE facilitates prediction of which areas to select and which categories within those areas to prioritize, based on its ability to judge long-term business trends (beyond short-term trends) from a market development perspective. This may include analysis of historical and current data (near real-time as needed) for one or more IP areas. A GAIE specializing in intellectual property monetization can connect investments and actions that have occurred in the intellectual property field, such as patent sales and licensing, with historical and / or current data. A GAIE trained in intellectual property monetization can also develop specific leads or domain categories with the highest probability of success based on past sales and / or licensing performance and / or market trends. While these decisions may involve risk, leveraging a trained GAIE can mitigate that risk and increase the predictability of decisions, especially in the future where corporate data is increasing and accessible through various channels.

[0304] GAIE can be configured, trained, and fine-tuned for a variety of functions, such as: proprietary data ingestion, routing, result determination, data publication / access authorization, predictive generation, and pattern recognition. Another example of a fine-tuned GAIE application could include hierarchical voice and visual commands. This could involve gradually changing the tone, volume, and intervals, similar to avionics, to generate scripts for narrating data or presentation materials. This would enable the development of synthetic speech technology that produces realistic (AI-generated) voices for podcasts, slideshows, and professional presentations. This could potentially eliminate the need for voice actors or complex recording equipment (e.g., background noise separation, dubbing, etc.).

[0305] As a concrete example, the GAIE system can facilitate news distribution using NPC-type avatars and adapt existing "clickbait" content to convey global events and situations in a conversational format. In this example, a news-based GAIE conversational agent is placed in the metaverse environment to notify users of recent events in a conversational format.

[0306] Furthermore, within the context of metaverse technology, generative AI conversation agents can be configured to constitute a metaverse.

[0307] Furthermore, in the context of metaverse technology, GAIE systems can collect information from real-world sensors, enabling them to augment the training data of customized conversational agents with real-time sensor datasets. For example, a training data augmentation system may be configured to augment the training of conversational agents using data from real-time sensor datasets. In addition, metaverse-related GAIE systems enable the augmentation of the training data of customized conversational agents using process result data. A training data augmentation system may be configured to augment the training of conversational agents with process result data from process result datasets, user behavior data, etc. In an exemplary embodiment, a GAIE-based training data augmentation system may be enabled (e.g., pre-trained) to augment the training of conversational agents with user behavior data from user behavior datasets.

[0308] In the examples, the application of a finely tuned GAIE system in the area of ​​governance can facilitate the advancement of governance automation, such as managing the use of copyrighted material. GAIE-based governance systems can further enhance AI training governance, including conversational AI training dataset management for bias and error correction, and conversational AI management for contextual appropriateness and other stylistic requirements. Finely tuned GAIE systems can further improve confidentiality governance, such as the gradual application of secret, proprietary, and confidential information elements permitted based on the depth of the conversation. Governance is also applicable to individuals. Therefore, a governance-fine tuned GAIE system can enhance and / or automate the determination of trustworthiness metrics for users who may interact with generative conversational AI systems. Furthermore, a governance-tuned GAIE system can enhance governance for generative AI systems, such as determining trustworthiness evaluation criteria for generative conversational AI systems. In general, governance use cases can be further expanded based on GAIE's topic-specific training capabilities.

[0309] A finely tuned GAIE system can play a role in identifying and managing systemic risks and discovering opportunities. A GAIE-based risk identification system responds to risk-related questions such as "What else should we know and pay attention to?" by curating datasets and automating the systemic risk identification process. Specifically, it identifies high-probability scenarios and the risks and opportunities arising from them, clarifies solution pathways, and recommends solutions.

[0310] For example, a GAIE-based risk identification system could have provided market participants and regulators with the insight that "U.S. banks are holding over $600 billion in unrealized Treasury bond losses" in response to the above instructions. Furthermore, the system may have provided specific information about banks that have become major outliers, posing significant systemic risk due to their scale and concentration. This system can be configured to issue system-wide warnings, preventing worst-case scenarios not only at outliers but across the entire risk pool. Specifically, risk-responsive GAIE systems capable of identifying hidden and unrecognized risks are applicable to areas outside of finance. However, even within the financial sector, such finely tuned GAIEs, with sufficient context, can highlight these hidden and underrecognized risks and provide options for resolving these excessive risks.

[0311] Another area of ​​risk identification and / or management concerns is the security implications associated with GAIE systems configured to generate computer-executable code. At the very least, relying on computers to write code raises questions about what security measures are effective and which can be circumvented by AI.

[0312] Furthermore, the areas of risk identification, management, and opportunity creation can also be applied to copyrighted material. Automated code generation may unintentionally introduce copyrighted material such as algorithms. Risk-focused copyright GAIEs can help detect potential copyright infringements in any program code, including machine-generated code.

[0313] Risk identification of visual training sets (images, graphs, etc.) is enhanced by a finely tuned GAIE capable of processing authenticity indices encoded as non-visual data. This is analogous to a tail voltage device that adds messages to the ends of a sine wave. Visual training sets can be encoded with non-visual authenticity indices detectable by the finely tuned GAIE.

[0314] Another risk identification-related area is fraud detection. Integrating customer fraud reports and inquiries into pre-trained data can enhance comprehensive scoring, including composite scores that integrate customer evidence, transaction data, and environmental trends. For example, an AI-based fraud detection system can achieve a comprehensive fraud detection approach by integrating customer fraud reports and inquiries into training / inquiry datasets and building a comprehensive scoring system that utilizes composite scores combining customer evidence, transaction data, and environmental trends.

[0315] Image processing applications can benefit from a finely tuned GAIE system. In an exemplary embodiment, optical content (e.g., screenshots) is processed by a machine vision system, and GAIE is enabled to describe the scene within the optical content using a generative conversational AI agent. In an exemplary embodiment, GAIE may be configured as the first AI / NN subsystem in a dual-process artificial neural network (DPANN) architecture. Such a DPANN architecture may include a formal logic-based and / or fuzzy-based system as the second NN subsystem. These DPANN systems implement the learning process, model management, etc. In an exemplary embodiment, the DPANN architecture may include capabilities to describe the construction and management of large-scale models.

[0316] Referring to Figure 8, the platform 800 for applying generative AI may include a robust task-independent next-token prediction AI engine 802 that predicts the next token given an input set encoded as embedded tokens. The robust task-independent next-token prediction AI engine 802 may include deep learning models that use multi-layer neural networks to process, analyze, and predict complex data such as language. The objectives of the robust next-token prediction AI engine 802 include data science modeling through the use of topic-specific embeddings, attention mechanisms, and decoder-specific transformer models. Capabilities of such an engine 802 include pre-training capabilities to facilitate next-token prediction settings for specific subjects (e.g., marketplace product ratings), tokenization capabilities to convert complex terminology into actionable tokens (e.g., converting complex chemical names to basic elements), access to distributed learning (e.g., data-parallel learning and / or model-parallel learning), and short-shot learning to reduce training demands for updates (e.g., new business intelligence data). Generally, the next token prediction AI engine 802 combines large-scale language modeling techniques with a dedicated transformer model for decoders to generate a robust foundational model for next token prediction AI content generation.

[0317] In the embodiment, the next token prediction AI engine 802 may consist of a machine learning (sparse multilayer perceptron) architecture configured to sparsely activate conditional computations using, for example, a mixture of experts (MoE) technique. The machine learning architecture consists of expert modules used to process the input and gate functions that facilitate the assignment of expert modules to process a portion of the input tokens. Furthermore, it includes configurations that combine deterministic routing of input tokens with trained routing that uses a portion of the input tokens to predict expert modules for a set of input tokens.

[0318] GAIEs are trained to operate within domains such as written languages, computer programming languages, or specific subject areas (e.g., the marketplace domain that software controls), and generate content (constructs) that conform to the rules of that domain. Generally, a GAIE can generate content on any subject it is trained on. For example, a GAIE trained on the subject of pig farming can generate language-based descriptions, images, contracts, breeding guidance, text output, and more for any of the broad subtopics related to pig farming.

[0319] Adapting a generative AI engine to a subject-specific application involves pre-training a next-token prediction AI model-based system using, for example, contextual (application, domain, subject-specific, etc.) examples that respond to corresponding prompts. Even if the predictive capability of the underlying next-token prediction AI engine itself is not affected by pre-training, it is possible to develop and deploy pre-trained instances for a specific subject.

[0320] In the embodiment, the platform 800 for applying generative AI may include a set of subject-specific pre-trained examples and prompts 804. This set of examples and prompts 804 consists of analyzing information characterizing various aspects of a domain (e.g., by human experts and / or computer-based experts and / or digital twins) to generate example prompts and preferred and / or correct responses. Pre-training also includes training the next token prediction AI engine 802 by sampling text (e.g., prompt / response sets) from the set of subject-specific pre-trained examples and prompts 804 and training it to predict the next word, object, and / or term. Pre-training also includes sampling images, contracts, blueprints, etc., to predict the next token. These prompt-response subsets facilitate the pre-training of the prediction AI engine 802 to predict the next token (e.g., word, object, image element, etc.) in various aspects.

[0321] When an instance is implemented for text generation, such a GAIE instance may be referred to as a natural language generation system that constructs words (e.g., from subword tokens), sentences, and paragraphs for a subject and / or domain.

[0322] In the embodiment, real-world instances of Platform 800 may require continuous updates to ensure Platform 800 remains responsive as aspects of the domain (e.g., entities within the domain) change (e.g., changes in business objectives, new product releases, mergers with competitors, emergence of new markets). In this regard, automating and iterating the process of training Platform 800 with contextually relevant prompts and examples each time new data is released to the enterprise can prevent time-dependent errors based on snapshot data at a specific point in time. The Generative AI-Applied Platform 800 may include a Continuous Pre-Training Module 828 that translates new and updated content into prompt and / or response sets and interactively executes the pre-training iteration process. New and updated data and information can be periodically obtained from various domain-specific information sets, such as: medical record datasets (e.g., medical diagnostic support), legal document and court decision datasets (e.g., legal advice), new product releases (e.g., product images), and financial datasets such as SEC filings and analyst reports. In the examples, using Platform 800 involves applying pre-training and optimization techniques to various fields (e.g., medical diagnostics, business operations, marketplace operations, etc.) to generate a finely tuned, domain-specific token prediction engine, including continuous improvement through (daily) in-context pre-training.

[0323] In an exemplary embodiment, the ongoing pre-training module 828 may work in conjunction with the next token prediction AI engine 802 to update the set of subject-specific tokens held in the subject-specific instance token storage device 808. This subject-specific instant token storage device 808 is referenced by the subject-specific instance of the next token prediction AI engine 802 during operation (e.g., during input / prompt processing). In an exemplary embodiment, the platform 800 may include multiple sets of subject-specific tokens maintained by the corresponding continuous pre-training module 828.

[0324] However, training does not guarantee that responses to prompts will always be accurate. Generally, business entities will not be very interested in tools that provide answers that are probably correct but may occasionally be inaccurate. Products that can provide accurate responses (e.g., including taking actions) based on end-user requests significantly expand potential use cases and product value. With high accuracy and integration with operational systems, such tools can go beyond mere new content generation and contribute to increased productivity. Through integration with workflows, they can facilitate the automation of workflow actions. From this perspective, the generative AI application platform 800 may include a pre-training optimization engine 806 that works in conjunction with the continuous pre-training module 828 to further improve domain-specific prompt response accuracy. The pre-training optimization engine 806 facilitates contextually accurate responses, task-specific fine-tuning, and, in sparse model variants of platform 800, enhances the learning ability with small amounts of training data. In the embodiment, fine-tuning brings further benefits to the platform by reducing any bias that may be present in the training data. This is essential to ensure that domain-specific terminology is adapted as training data changes (e.g., ensuring that "influencer" is replaced with "creator" in the digital marketing / promotion field). Furthermore, the pre-trained optimization engine 806 enhances the platform's user-centric response capabilities by providing a wider range of prompts and responses based on user preferences (e.g., speaking style). In exemplary embodiments, user-centric responses include fine-tuning platform 800 for different roles within an organization. For example, if a financial planner inquires about a business development topic, the response would be tailored to the financial planner's role (e.g., compared to an inquiry on the same topic from a customer / client).

[0325] Platform 800, to which generative AI is applied, is used to generate text-based content for multinational corporations with employees who speak different languages. While Platform 800 may be trained (including pre-training) to operate interactively in multiple languages, the use of the neural machine translation module 810 is beneficial for creating automatically generated content. In the embodiment, some organizations within a first jurisdiction generate content in their first language, and the resulting iteratively generated output (e.g., various reports) may be in their first language. However, for employees who speak a second language, it may be beneficial to have these reports translated into their native language. Therefore, associating the neural machine translation module 810 with Platform 800 while reducing its computational load could be a valuable approach.

[0326] Emerging next-token prediction AI systems feature highly adaptable next-token prediction capabilities. These capabilities can be further adapted to support solution prediction for closed problem sets, such as resource allocation or robot fleet deployment. To achieve more advanced predictive capabilities, subject-specific next-token prediction AI engines, such as the Generative AI Application Platform 800, may include a "solution prediction engine 812" that predicts the most promising solution to a closed solution set problem. This leverages next-token (e.g., next-word) prediction capabilities. This can be achieved by optionally using a set of pre-trained prompts and examples specific to the problem domain. These examples can be adapted to different user preferences. In an embodiment, each user in a closed problem set environment generates prompts and responses, and the Platform 800 responds based on the user's inquiry style. Alternatively, the solution prediction engine 812 attempts to provide responses consistent with user preferences (e.g...

Claims

1. A computer-implemented system that provides an embedded marketplace within a host application, A data classification module configured to classify data into classified data based on predefined confidentiality levels and regulatory compliance requirements, An access control module configured to manage permissions for different user roles within an enterprise and to grant access to the classified data in accordance with the confidentiality level and regulatory compliance requirements, A data format module configured to convert the aforementioned classified data into formatted data in a display format customized for various corporate departments, An integrated module configured to interface with at least one of an Enterprise Resource Planning (ERP) system or a Customer Relationship Management (CRM) system, and to acquire and classify the said data, A system comprising a user interface module configured to present the formatted data within the host application and to provide a seamless user experience in accessing the embedded marketplace.

2. The system according to claim 1, wherein the host application for embedding the marketplace is an enterprise resource planning (ERP) system, and the data classification module is further configured to classify financial data, supply chain data, and human resource data for selective presentation to authorized users.

3. The system according to claim 1, wherein the host application for embedding the marketplace is a customer relationship management (CRM) system, and the data format module is further configured to generate visual sales funnels and marketing campaign analyses for sales and marketing departments.

4. The system according to claim 1, wherein the host application for embedding the marketplace is a product lifecycle management (PLM) system, and the integration module is further configured to provide research and development data, including product specifications and test results, formatted as technical documentation.

5. The system according to claim 1, wherein the host application for embedding the marketplace is a governance, risk, and compliance (GRC) platform, and the access control module is further configured to enforce compliance with legal and regulatory standards by restricting access to compliance-related sensitive data.

6. The system according to claim 1, wherein the host application for embedding the marketplace is an IT service management tool, and the user interface module is further configured to display IT asset management data, system performance indicators, and security incident reports in a format suitable for use by the IT department.

7. The system according to claim 1, wherein the host application for embedding the marketplace is an intranet portal within the company, and the data format module is further configured to provide management dashboards, departmental reports, and company-wide announcements in a centrally managed location.

8. The system according to claim 1, wherein the host application for embedding the marketplace is a cloud-based collaboration platform, and the integration module is further configured to facilitate data sharing and project management among geographically dispersed teams within the enterprise.

9. A computer-implemented system for managing an embedded marketplace within an enterprise, A data classification module configured to classify corporate data into classified data based on predefined confidentiality levels and regulatory compliance requirements, An access control module configured to manage permissions for different user roles within the said company and to grant access to the classified data in accordance with the said confidentiality level and the said regulatory compliance requirements, A data formatting module configured to convert the aforementioned classified data into formatted data in a display format customized for a series of corporate departments, An integrated module configured to interface with an enterprise system and to acquire and classify said enterprise data, wherein the enterprise system includes at least one of an enterprise resource planning (ERP) system or a customer relationship management (CRM) system, A system comprising a user interface module configured to present the formatted data within the host application in order to access the embedded marketplace.

10. The system according to claim 9, characterized in that the data classification module utilizes role-based access control (RBAC) to assign data access privileges.

11. The system according to claim 9, characterized in that the data classification module tags the data with metadata indicating its confidentiality level.

12. The system according to claim 9, characterized in that the access control module includes functions for periodic auditing and real-time monitoring of data access.

13. The system according to claim 9, characterized in that the data format module provides management summaries including high-level graphics, dashboards, and overviews for management.

14. The system according to claim 9, characterized in that the data format module provides detailed reports, raw datasets, and analytical tools for detailed data analysis by employees.

15. The system according to claim 9, characterized in that the user interface module enables a data view that can be customized according to the needs of the department.

16. The system according to claim 9, characterized in that the integrated module includes a data services catalog equipped with tools for processing, analyzing, and visualizing data.

17. The system according to claim 9, characterized in that the integrated module is configured to extract relevant data from CRM, SCM, and PLM systems and format it for different departments.

18. The system according to claim 9, characterized in that the user interface module provides training modules and support services to assist employees in utilizing data.

19. The system according to claim 9, characterized in that the integrated module provides a unified platform for centralized data governance across the entire enterprise.

20. The system according to claim 9, characterized in that the access control module supports attribute-based access control (ABAC) and RBAC.

21. The system according to claim 9, characterized in that the integrated module automates compliance with regulations by directly incorporating rules into the data access mechanism.

22. The system according to claim 9, characterized in that the user interface module provides advanced search functions to improve data discoverability.

23. The system according to claim 9, characterized in that the user interface module provides personalized data and service recommendations based on user roles and past usage.

24. The system according to claim 9, wherein the integration module is configured to dynamically adjust permissions based on context, and the context includes at least one of the current project or collaboration.

25. The system according to claim 9, characterized in that the integrated module supports a scalable architecture to accommodate the increasing volume and diversity of data.

26. The system according to claim 9, characterized in that the user interface module provides an intuitive interface that eases the user's learning curve.

27. The system according to claim 9, characterized in that the integrated module includes usage tracking and analysis functions that provide insights into data value and usage patterns.

28. The system according to claim 9, characterized in that the integrated module maintains a comprehensive audit trail for security audits and compliance checks.

29. The system according to claim 9, characterized in that the integrated module facilitates subscription-based access to data services for predictable budgeting and cost management.

30. A computer-implemented method for integrating a system into a host platform for process automation and artificial intelligence, The processing system includes the steps of identifying a set of functions provided by the host platform, The processing system includes the steps of determining a set of marketplace services related to the identified function of the host platform, The processing system comprises the steps of integrating the interface of the marketplace service into the host platform, wherein the interface is configured to provide the marketplace service contextually based on user interaction with the host platform. The processing system comprises the steps of configuring the marketplace service to personalize the marketplace service provided to the user using data from the host platform, A method characterized by comprising the step of enabling the user to perform transactions within an embedded marketplace without leaving the host platform, using the processing system.

31. The method according to claim 30, wherein the host platform includes an enterprise resource planning (ERP) system, and the marketplace services are selected based on procurement needs identified by the ERP system.

32. The method according to claim 30, wherein the host platform includes a customer relationship management (CRM) system, and the marketplace service is configured to provide products or services based on customer profiles and interactions stored within the CRM system.

33. The method according to claim 30, wherein the host platform includes a social media platform, and the marketplace service is configured to provide products or services related to content viewed by the user on the social media platform.

34. The method according to claim 30, characterized in that the host platform includes an Internet of Things (IoT) device, and the marketplace service is configured to provide maintenance, repair, or related products based on sensor data collected by the IoT device.

35. The method according to claim 30, wherein the host platform includes a digital wallet application, and the marketplace service is configured to provide financial products or services based on the user's financial transactions and preferences.

36. The method according to claim 30, wherein the host platform includes a content creation platform, and the marketplace service is configured to provide digital assets, tools, or services related to the content created by the user.

37. The method according to claim 30, wherein the host platform includes a gaming platform, and the marketplace service is configured to provide in-game items, virtual goods, or physical goods related to the game being played by the user.

38. The method according to claim 30, characterized in that the marketplace service includes an artificial intelligence algorithm for predicting user needs and proactively presenting relevant marketplace services within the host platform.

39. The method according to claim 30, wherein the marketplace service is configured to utilize process automation for processing transactions within the embedded marketplace, and the transactions include settlement processing, order fulfillment, and post-transaction customer service.

40. The processing system includes the steps of collecting user feedback regarding the marketplace service, The method according to claim 30, further comprising the step of using an artificial intelligence algorithm by the processing system to adjust the marketplace service based on the collected feedback and improve relevance and user satisfaction within the embedded marketplace.

41. A computer-implemented method for managing procurement within a company, Steps to intercept web browser communications initiated by company employees, The steps include analyzing the intercepted communications and identifying procurement-related activities, The steps include accessing a regulatory database and determining compliance with applicable laws and corporate policies, Steps include evaluating procurement requests based on budget constraints and employee approval levels, A method comprising the step of controlling the execution of a procurement transaction by allowing, modifying, or blocking it based on compliance and approval evaluations.

42. A system for automated procurement management in a corporate environment, A network communications analysis module configured to monitor and evaluate web-based procurement activities, A compliance assessment engine integrated with a regulatory database for real-time compliance verification, An approval management module that facilitates and tracks the procurement request approval process, A system characterized by including a transaction execution module that enforces compliance and approval results by managing the confirmation of procurement transactions.

43. A computer-implemented system for integrating a marketplace into a digital twin, A processing system configured to generate a digital twin representing a physical asset, wherein the digital twin includes real-time data that reflects the state, status, and performance of the physical asset. A marketplace module configured to facilitate transactions related to the physical assets and incorporated within the digital twin, the marketplace module includes listing functions, purchasing functions, and transaction processing functions, A data analysis module configured to identify transaction needs or opportunities within the marketplace module using the real-time data from the digital twin, A system comprising a communication interface configured to present transaction opportunities to a user and to enable the user to interact with the marketplace module via the digital twin.

44. The system according to claim 43, wherein the marketplace module is further configured to provide predictive maintenance services for the physical assets based on the analysis of the real-time data by the data analysis module.

45. The system according to claim 43, wherein the marketplace module is further configured to provide recommendations for spare parts and consumables compatible with the physical assets.

46. The system according to claim 43, wherein the marketplace module includes a smart contract function configured to automate the execution of transactions based on predetermined rules derived from the real-time data.

47. The system according to claim 43, wherein the marketplace module is further configured to provide insurance services, and the terms of the insurance services are dynamically adjusted based on the real-time data from the digital twin.

48. The system according to claim 43, wherein the marketplace module is further configured to facilitate the resale or lease of the physical asset by connecting potential buyers or lessees with the digital twin.

49. The system according to claim 43, further characterized in that the marketplace module is configured to cooperate with third-party service providers, enabling the provision of extended services related to the physical assets.

50. The system according to claim 43, further characterized in that the marketplace module is configured to utilize a machine learning algorithm to personalize the transaction opportunities presented to the user based on the user's behavior and preferences.

51. The system according to claim 43, wherein the marketplace module is further configured to support a virtual reality interface, enabling the user to interact with the digital twin and the marketplace in an immersive environment.

52. The system according to claim 43, wherein the marketplace module is further configured to provide a platform for user-generated content, allowing users to post custom modifications or improvements related to the physical assets.

53. The system according to claim 43, further characterized in that the marketplace module is configured to aggregate data from multiple digital twins representing a fleet of similar physical assets and to provide opportunities for bulk transactions.

54. The system according to claim 43, wherein the marketplace module is further configured to enable energy trading services based on real-time energy usage and production data for digital twins representing energy consumption assets or energy generation assets.

55. The system according to claim 43, wherein the marketplace module is further configured to provide subscription-based services related to the physical assets, and the subscription conditions are changeable in response to changes in the real-time data.

56. The system according to claim 43, further characterized in that the marketplace module is configured to provide a feedback mechanism for users to evaluate and review transactions and to adapt the presentation of future transaction opportunities.

57. The system according to claim 43, wherein the marketplace module is further configured to support regulatory compliance monitoring, and transactions are automatically adjusted to comply with applicable laws and regulations based on the real-time data.

58. A system that provides an integrated transaction platform, An embedded marketplace module configured to aggregate offerings from multiple vendors within the user interface of a host application, A data aggregation system configured to collect and process data from various sources and personalize the aggregated offerings based on user preferences and behavior, A transaction execution module configured to facilitate the purchase, sale, and exchange of goods and services within an embedded marketplace, A blockchain interface configured to work with one or more distributed ledgers to record transactions performed within the aforementioned embedded marketplace, A system comprising a smart contract module configured to generate and execute contracts related to transactions within the embedded marketplace based on predetermined rules and conditions.

59. The system according to claim 58, wherein the embedded marketplace module is further configured to present a unified view of the aggregated offerings across multiple external marketplaces.

60. The system according to claim 58, characterized in that the data aggregation system utilizes machine learning algorithms to refine personalization based on real-time user interaction with the embedded marketplace.

61. The system according to claim 58, wherein the transaction execution module is further configured to process payments using at least one of fiat currency or cryptocurrency.

62. The system according to claim 58, wherein the blockchain interface is further configured to support multiple blockchain protocols in order to ensure compatibility with various distributed ledger technologies.

63. The system according to claim 58, further characterized in that the smart contract module is configured to automatically adjust the contract terms based on changes in regulatory requirements.

64. The system according to claim 58, further comprising a robotic process automation (RPA) module configured to automate the procurement process based on inventory levels and forecast demand analysis.

65. The system according to claim 64, further characterized in that the RPA module is configured to work in conjunction with a vendor management system in order to streamline supply chain operations.

66. The system according to claim 58, further characterized in that the embedded marketplace module is configured to work in conjunction with a digital twin representation of physical assets to enhance the visualization of offerings.

67. The system according to claim 58, wherein the transaction execution module includes a recommendation engine for suggesting ancillary services related to the main offering.

68. The system according to claim 58, characterized in that the blockchain interface is configured to tokenize assets in order to facilitate asset trading within the embedded marketplace.

69. The system according to claim 58, characterized in that the smart contract module includes a dispute resolution mechanism that operates automatically based on transaction anomalies.

70. The system according to claim 58, further characterized in that the data aggregation system is configured to aggregate at least one of social media data or IoT device data in order to enhance the personalization of the offerings.

71. The system according to claim 58, wherein the embedded marketplace module is further configured to provide location information services and to provide goods and services related to the user's geographical location.

72. The system according to claim 58, wherein the transaction execution module is further configured to support subscription-based transactions for recurring purchases within the embedded marketplace.

73. The system according to claim 58, wherein the blockchain interface is further configured to provide an audit trail of transactions in order to ensure transparency and compliance.

74. The system according to claim 58, further characterized in that the smart contract module is configured to cooperate with an external contract management system for contract synchronization between platforms.

75. Furthermore, the system according to claim 58 is characterized by including an RPA module configured to automate compliance checks against corporate policies during the procurement process.

76. The system according to claim 58, wherein the embedded marketplace module is further configured to incorporate a marketplace into a virtual reality environment for an immersive shopping experience.

77. The system according to claim 58, further characterized in that the transaction execution module is configured to utilize the blockchain interface and the smart contract module to enable peer-to-peer transactions without the involvement of an intermediary.

78. A computer-implemented system for managing transactions within an enterprise ecosystem, It includes a processor and memory for storing instructions, When the aforementioned instruction is executed by the processor, it is directed to the system, Integrating an embedded marketplace with an Enterprise Access Layer (EAL) that interfaces with multiple enterprise resources, By integrating the aforementioned embedded marketplace with the company's workflow system, the procurement and sales processes can be automated. Using a data service system, manage listing information, transactions, and user profiles within the embedded marketplace. To implement an intelligence system that provides predictive analytics for market trends and demand forecasting within the aforementioned embedded marketplace, Enforce security and compliance through an authorization system that controls access to the functions of the aforementioned embedded marketplace. Managing digital transactions through a wallet system that works in conjunction with the aforementioned embedded marketplace, and A system characterized by generating reports on marketplace activities through a reporting system that communicates with the aforementioned embedded marketplace.

79. The system according to claim 78, wherein the instruction further causes the system to collect real-time data, analyze the real-time data, and provide the user with personalized recommendations regarding goods or services based on the user's past transaction data.

80. The system according to claim 78, further characterized in that the instruction causes the system to implement a smart contract orchestration engine to automate transactional workflows within the enterprise ecosystem.

81. The system according to claim 78, further characterized in that the instruction causes the system to operate in conjunction with technologies deployed on the company's private network, which include on-premises cloud resources and platforms, or at least one of the cloud resources and platforms.

82. The system according to claim 78, further characterized in that the instruction causes the system to tokenize digital assets and to digitally represent transactions within the enterprise ecosystem.

83. The system according to claim 78, further characterized in that the instruction causes the system to facilitate transactions with external entities by providing a set of network resources for two-party or multi-party transactions involving the enterprise.

84. The system according to claim 78, further characterized in that the instruction causes the system to simplify transactions for a company by enabling the company to interface with multiple markets, marketplaces, exchanges, and platforms through a common access point.

85. The system according to claim 78, further characterized in that the instruction causes the system to employ blockchain to manage and protect transactions within the enterprise ecosystem.

86. The system according to claim 78, further characterized in that the instruction causes the system to include a generative content system that proposes new workflows for enterprise processes using a large-scale language model (LLM) trained on enterprise-specific data.

87. A computer-implemented system for facilitating transactions within an embedded marketplace enterprise ecosystem, It includes a processor and memory for storing instructions, When the aforementioned instruction is executed by the processor, it is directed to the system, Integrating an embedded marketplace with a company's digital infrastructure, By linking the aforementioned embedded marketplace with the company's workflow system, the transaction process can be automated. Using a data service system, manage listing information, transactions, and user profiles within the embedded marketplace. Through the intelligence system, provide analysis and insights for strategic decision-making within the embedded marketplace. Enforcing security and compliance protocols through an authorization system that controls access to the aforementioned embedded marketplace, and A system characterized by facilitating digital transactions through a wallet system linked to the aforementioned embedded marketplace.

88. The system according to claim 87, characterized in that the embedded marketplace utilizes a large-scale language model (LLM) trained on enterprise-specific data to assist in the generation and optimization of transaction workflows.

89. The system according to claim 87, characterized in that the embedded marketplace employs robotic process automation (RPA) to streamline procurement and sales processes by automating repetitive tasks and data processing.

90. The system according to claim 87, wherein the embedded marketplace includes a digital twin of the enterprise ecosystem for simulating and analyzing marketplace dynamics and enterprise resource planning scenarios.

91. The system according to claim 87, characterized in that the embedded marketplace works in conjunction with a blockchain network to manage and protect transactions and ensure data integrity and traceability.

92. The system according to claim 87, characterized in that the embedded marketplace is configured to use artificial intelligence (AI) for a dynamic pricing strategy based on real-time market data and predictive analytics.

93. The system according to claim 87, characterized in that the embedded marketplace incorporates a smart contract orchestration engine for automating the execution of agreements and compliance with contract terms.

94. The system according to 87, characterized in that the embedded marketplace is interfaceable with Internet of Things (IoT) devices and facilitates transactions based on sensor data and automatic triggers.

95. The system according to claim 87, characterized in that the embedded marketplace uses machine learning algorithms to personalize product recommendations based on user behavior and preferences.

96. The system according to claim 87, wherein the embedded marketplace is further configured to support subscription services that enable recurring transactions and customer retention strategies.

97. The system according to claim 87, wherein the embedded marketplace includes a virtual assistant driven by natural language processing (NLP) to help users navigate the marketplace and complete transactions.

98. The system according to claim 87, characterized in that the embedded marketplace is designed to work in conjunction with a virtual reality and augmented reality (VR / AR) platform to provide immersive product demonstrations and virtual showrooms.

99. The system according to claim 87, characterized in that the embedded marketplace is configured to tokenize digital assets representing ownership and transactions of digital and physical goods within the enterprise ecosystem.

100. The system according to claim 87, wherein the embedded marketplace is further configured to facilitate cross-border transactions by incorporating multi-currency and multi-language support.

101. The system according to claim 87, characterized in that the embedded marketplace improves customer service and engagement by employing a customer relationship management (CRM) system to track interactions and transactions with customers.

102. The system according to claim 87, further characterized in that the embedded marketplace is configured to work in conjunction with a supply chain management system to optimize inventory levels and logistics.

103. The system according to 87, characterized in that the embedded marketplace includes an API gateway that enables third-party applications and services to interact with the marketplace ecosystem.

104. The system according to claim 87, further characterized in that the embedded marketplace is configured to employ a fraud detection system that identifies and prevents fraudulent transactions using anomaly detection technology.

105. The system according to claim 87, characterized in that the embedded marketplace is configured to support a peer-to-peer (P2P) network for direct transactions between users without the involvement of an intermediary.

106. The system according to claim 87, characterized in that the embedded marketplace includes a feedback and evaluation system that employs sentiment analysis to measure customer satisfaction and improve service delivery.

Images