4216 results about "Language model" patented technology

Systems and methods for protecting and fortifying machine learning engines, artificial intelligence (AI) engines, large language models, deep learning engines, reinforcement learning engines, and AI-based agentic units. An Offline Protection Unit analyzes characteristics of a Protected Engine, and performs offline fortification of the Protected Engine against attacks; by changing operational properties or operational parameters of the Protected Engine to reduce its vulnerability to attacks. An Online Protection Unit performs analysis of at least one of: (i) inputs that are intended to be inputs of the Protected Engine, (ii) outputs that are generated by the Protected Engine; and based on the analysis, dynamically performs online fortification of the Protected Engine against attacks; by dynamically changing operational properties or operational parameters of the Protected Engine to reduce its vulnerability to attacks.

The invention discloses a knowledge base enhancement generation method and system based on hybrid retrieval and fact verification. The method comprises the following steps: receiving an original query of a user; performing intention analysis and rewriting on the query, identifying an intention type and generating a sub-query adapted to retrieval; executing mixed retrieval of vector retrieval, keyword retrieval and selectable knowledge graph retrieval based on the sub-query, and recalling related knowledge fragments; performing deduplication clustering, fact conflict recognition processing and correlation reordering on the knowledge fragments; according to the intention type and the reordered knowledge fragment, dynamically selecting a prompt template to construct an enhanced prompt; and sending the enhanced prompt into the large language model, and generating a target response with the reference source. According to the method, knowledge recall comprehensiveness is improved through mixed retrieval, knowledge reliability is ensured through fact verification, generation logicality is enhanced in combination with dynamic prompt construction, and response accuracy and credibility are improved.

The invention provides a multi-agent collaborative question and answer enhancement method and system based on heterogeneous data knowledge, and belongs to the technical field of information management. The planning optimization intelligent agent carries out structured processing on the input complex natural language problem; a multi-modal retrieval mechanism on the heterogeneous knowledge source is constructed based on the problem disassembly and entity recognition result, and a multi-path recall agent passes through a semantic matching model of a double-tower structure; the abstract extraction agent performs semantic fusion and information extraction on the text segments and the knowledge graph sub-graphs output by the multi-path recall module; logic verification and quality evaluation are carried out on the answers generated by the abstract extraction module by the reflection iteration agent; and by setting a threshold mechanism and combining importance weights of the sub-questions, performing scoring and reflection optimization on the generated answers by utilizing a large language model LLM (Language Language Model). The method can effectively cope with cross-domain and multi-level complex question and answer tasks, and has good expandability and intelligent level.

The present disclosure includes computer-implemented methods of guardrails for securely using large language models (LLMs). The method comprises monitoring user data flow using an application programming interface (API) and receiving an administrative policy from an administration communication interface. The method involves dynamically applying a plurality of LLM input inspectors to LLM input data. The application of the plurality of LLM input inspectors is based on the administration policy. The dynamic application of the plurality of LLM input inspectors is in sequence for latency optimization. The plurality of LLM input inspectors serve as LLM input guardrails for a plurality of secure deployed large language models (LLMs). The plurality of LLM input inspectors are configured by the administrative policy and validate the LLM input data to validated LLM input data based on the administration policy. Additionally, the method comprises dynamically applying a plurality of LLM output inspectors to LLM output data.

The systems and methods disclosed herein orchestrate task execution among autonomous (or semi-autonomous) AI agentic models (“agents”) using a gateway router that dynamically coordinates the agents based on prompt characteristics, user context, and / or real-time operational factors. Received inputs (e.g., prompts) are segmented into subcomponents (e.g., sub-queries), which are routed / mapped to candidate agents based on the output parameters of the subcomponent (e.g., performance thresholds, cost thresholds) and operational parameters (e.g., cost, performance metric values, user access restrictions, timing restrictions) of each agent. The gateway router maintains dynamic routing data structures for each agent that are continuously updated based on environmental stimuli (e.g., geo-political stimuli, sensor stimuli, agent stimuli). For example, the gateway router causes agents to dynamically switch between rule engines identified by the routing tables in response to detecting environmental stimuli. Responses from the candidate agents are aggregated into an output that is responsive to the input.

A database system integrates in-database machine learning (ML) models with in-database large language models (LLMs) or other generative artificial intelligence (AI) models that enable new applications. The database system receives one or more inferences from an ML model and provides an inference input to a retrieval agent of an object store. One or more vector stores represent a plurality of reference documents using semantic encodings. The retrieval agent performs a similarity search of the one or more vector stores to retrieve a set of passages from the plurality of reference documents based on similarity of encodings of the inference input and encodings of passages in the plurality of reference documents. The database system generates a linguistic prompt for an LLM having a context including the inferences and passages and applies the LLM to the linguistic prompt to generate a natural language explanation of the one or more inferences.

A SQL generation method based on a large language model, comprising: generating data definition language (DDL) prompt slot information and data example slot information on the basis of a configuration operation of a user on a SQL database; receiving a natural language query request input by the user, and transcribing the natural language query request to obtain question transcribing slot information; on the basis of the question transcribing slot information, the DDL prompt slot information and the data example slot information, acquiring complete prompt information matched with a large language model; and inputting the complete prompt information into the large language model to generate an executable SQL statement matched with the natural language query request.

A multimodal intelligent agent system for dynamic environmental monitoring and user-centered support, consisting of: a multimodal sensor module configured to continuously acquire environmental and behavioral data from multiple input modalities, including at least one visual sensor, at least one acoustic sensor, at least one environmental conditions sensor, and at least one proximity or motion detection sensor, each generating modality-specific data streams representing visual images, audio waveforms, physical environmental parameters, and motion signatures within a monitored environment; a data preprocessing and fusion subsystem that is operationally coupled with the multimodal sensor module and configured to normalize, temporally align, and transform the modality-specific data streams into high-dimensional feature embeddings using a variety of encoders, wherein the visual encoder uses convolutional or vision transformer architectures, the audio encoder uses a spectral-temporal feature extractor, and the sensor encoder transforms raw analog data into context vectors suitable for multimodal alignment; a multimodal processing unit consisting of a transformer-based large language model (LLM) trained on paired multimodal datasets and configured to perform semantic fusion, context abstraction, and inference across the aforementioned aligned multimodal feature embeddings to generate a contextual understanding of environmental and behavioral states; an adaptive agent controller coupled to the multimodal inference processing unit and configured to instantiate, manage, and terminate a variety of task-specific intelligent agents, each agent being a software unit configured to perform a specialized function selected from meeting summarization, behavioral analysis, misplaced object detection, or environmental anomaly identification, with the agents dynamically interacting with the inference engine to retrieve contextually relevant multimodal embeddings for task execution; a personalization and adaptive learning subsystem consisting of a user preference database and a neural memory structure configured to update and refine model parameters based on user-specific interaction history, thereby enabling personalized output generation, prioritization of recommendations, and long-term behavioral adaptation; and An output generation interface is operationally connected to the adaptive agent controller and configured to produce multimodal output in textual, visual, and auditory form. The interface is capable of displaying human-readable summaries, notifications, and visual reconstructions of identified entities or environmental states.

The invention discloses an AR navigation system and method based on a visual language model, and the method comprises the steps: firstly carrying out video data collection, and constructing a memory database; secondly, querying a navigation target most related to a natural language request of a user in the constructed memory database to obtain a current AR equipment pose and a target pose; and then solving the shortest path from the current pose to the target pose by using the current AR equipment pose and the target pose according to the point cloud map, and optimizing the path direction to finally obtain an optimized path. And finally, guiding a user to move along the planned path through view superposition path indication and voice prompt in AR equipment by utilizing the optimized path, and updating the point cloud map and the memory database. According to the invention, high-precision real-time positioning and sparse point cloud map construction can be realized only by camera input in indoor and outdoor complex environments with low GPS precision, accurate navigation is carried out, and the flexibility and intelligent level of navigation are improved.

The invention provides a computer document intelligent compliance detection system based on deep learning, and relates to the technical field of data processing, and the system comprises the steps: carrying out the topological structure analysis of an adjacent matrix through a graph neural network, so as to detect an abnormality, obtaining a structure deviation degree, and analyzing a decision path node sequence through a pre-training language model to generate semantic conflict features; fusing the structure deviation degree with the semantic conflict feature to generate a risk feature vector; generating an augmented rule set according to the risk feature vector, and updating the formalized rule base; updating parameters of the feature coding component through gradient back propagation; optimizing a differentiable logic layer judgment threshold value and the weight of an analysis module; and obtaining the updated feature coding component, the rule base version and the risk quantification parameter. According to the invention, the efficiency of document compliance detection is effectively improved.

The invention relates to the technical field of artificial intelligence and natural language processing, in particular to a key value cache compression and sparse attention calculation method and system for large language model reasoning, and the method comprises the steps: an offline calibration stage; the online reasoning stage comprises the following steps: a pre-filling step; an autoregression generation step: for each newly generated lexical element, projecting a current query vector Q and a key vector K in a key cache to a low-dimensional space to obtain Q'and K '; calculating an approximate attention score based on Q'and K ', and selecting an index I of the first k most relevant lexical elements which are ranked from high to low; and calculating an accurate attention score based on Q and K [I], and calculating with the value vector V [I] to obtain the output of the current lexical element. According to the scheme, the memory and calculation bottleneck of large model reasoning in a scene of long text sequence input are solved, and the method has the advantages of reducing video memory occupation and calculation complexity at the same time.

The present disclosure relates to systems and methods for enhancing the interaction between users and automated agents, such as digital assistants, by employing Large Language Models (LLMs) to infer the intent of spoken language. The invention involves continuously monitoring ambient audio, converting speech to text, and utilizing LLMs to determine whether spoken language is intended for the automated agent. A structured prompt, including the converted text and specific instructions, is sent to the LLM, which is fine-tuned to process domain-specific prompts. The LLM provides a structured output in a standardized format, indicating the user's intent. The system may involve multiple prompts to perform separate tasks, such as identifying intent and generating additional context-specific data. This approach facilitates a more natural and intuitive user experience by eliminating the need for wake words and allowing seamless conversational interaction with virtual assistants across various platforms and devices.

A system automatically evaluates requests for escalation of privileges in an organization. The system receives a Request for Escalation of Privileges (REP), in a natural language, that indicates a request from a Requesting User to elevate access privileges to a computerized organizational resource. The system automatically feeds the REP as input into a fine-tuned Large Language Model (LLM), which automatically performs an analysis of that REP, and automatically generates an LLM-based output indicating at least: a proposed minimal set of elevated permissions that are estimated by the LLM to be needed and also sufficient for achieving a task described in the REP. The LLM is configured to further generate output with textual reasoning for inclusion of at least one elevated permission in that proposed minimal set of elevated permissions.

The invention discloses a safety control method based on a knowledge graph, and belongs to the field of safety control, and the method comprises the following steps: crawling various types of data according to keywords through a crawler program to construct a fireproof database; carrying out semantic analysis on data in the fireproof database by using a natural language processing technology, and carrying out data processing on the fireproof database after semantic analysis through word vector modeling and entity relationship extraction; constructing an internal large-scale language model by using the processed fireproof database, and performing hierarchical clustering on the internal large-scale language model by adopting a Leiden technology; establishing a multi-dimensional mapping model based on historic building spatial features, cultural relic features and disaster-inducing factors in a hierarchical clustering result, and constructing a multi-modal dynamic association fireproof knowledge graph by using a knowledge graph technology based on the multi-dimensional mapping model; and dynamically selecting a search mode based on a user query keyword, and generating a security control result according to the query keyword based on the search mode.

The invention relates to the technical field of machine vision, and discloses a machine vision-oriented cross-modal image-text analysis method, which comprises the following steps of: partitioning an input image to generate an image block sequence; inputting the image block sequence into a visual converter for multi-scale feature extraction, and generating target visual features; encoding the input text to generate a target text feature; inputting the target visual features and the target text features into a deep reconstruction bottleneck network for compression alignment, and generating a cross-modal compression vector; and inputting the cross-modal compression vector into a large language model to generate cross-modal decoding information, so that cross-modal redundant information can be effectively filtered, compact shared semantic representation can be learned, the information integrity of the compression process is ensured through bidirectional reconstruction verification, cross-modal semantic alignment is realized, and the method has the advantages of high efficiency and high reliability. Omnibearing cross-modal content generation from the whole to details is achieved, and the requirements of different application scenes are met.

Systems and methods for large language model (LLM)-agnostic data anonymization. Data anonymization includes data obfuscation (and data deobfuscation) to protect confidential information a user is going to send to an LLM service or application programming interface (API). Encryption can be used for data obfuscation and particularly, for securing data from unauthorized access. Likewise, decryption can be used for data de-obfuscation.

The invention relates to the technical field of artificial intelligence, can be applied to business scenes of financial science and technology, medical health and the like, and discloses a machine learning task execution and feedback method, device, equipment and medium. And performing semantic reasoning by using a pre-training language model, generating a decision result in combination with an entity relationship associated with a task type and a data type in the knowledge graph, executing a corresponding machine learning task, converting a task execution process and an execution result into natural language feedback information, and outputting the natural language feedback information to a user side. According to the method, natural language input is structured into machine learning task description, large model semantic reasoning and knowledge graph knowledge association are combined, an executable decision result is generated, automatic closed loop from task recognition, model selection to result feedback is achieved, the user operation threshold is lowered, and the model development efficiency and the interaction intelligence level are improved.

A computer-implemented method generates workflow definitions in workflow definition language using one or more large language models, LLMs. The method includes receiving a natural language description of an automated workflow and generating a plan generation prompt including the natural language description and plan generation instructions. The plan generation prompt is input to one of the LLMs and in response a structured plan comprising a plurality of actions are received. For each action, a corresponding segment of workflow definition language is generated to provide a plurality of segments of workflow definition language. The segments are combined to form a workflow definition corresponding to the natural language description.

Techniques for cache management for reducing latency in LLM inferencing are described. In some embodiments, a system caches encoded data of portions of a prompt so that the encoded data is available for use by the LLM across dialog turns of a dialog session. Within a dialog session, a portion of the LLM prompt may be the same across dialog turns, and instead of recomputing the attention / encodings for such portions, the cached encodings can be used by the LLM during processing. In some embodiments, user inputs for the dialog session may be routed to the same LLM container and encoded data for the dialog session may be stored at the same cache associated with the LLM. In some embodiments, the system enables asynchronous prompt encoding while performing ASR processing.

Systems and methods for detecting and mitigating prompt injection attacks on a generative LLM are disclosed. A deployment scenario is considered, in which the generative LLM supports a task automation function. Prompts are received and interpreted by the generative LLM, and outputs from the generative LLM are used to trigger automation actions. The prompts are constructed based on a combination of user input and external data and are, therefore, vulnerable to prompt injection attacks though manipulation of the external data. To mitigate this risk, a separate discriminative classification, decoupled from the generative LLM, engine is configured to identify malicious prompts, and filter out any malicious prompts before they reach the generative LLM.

Techniques for updating a large language model (LLM) to correct generation of undesired responses, such as incorrect outputs, toxic outputs, etc. are described. Typical methods of retraining and fine-tuning are inefficient and computationally expensive for LLMs. Some embodiments of the present disclosure involve identifying a salient layer of the LLM that is responsible for the undesired response and editing only the salient layer. This layer is identified by computing a saliency value for the layer using a mean of gradient values for the layer, and the layer with the greatest saliency value is selected for editing. For editing, a small network is used to update the weights of the selected layer. The LLM is updated to include the edited layer, and the updated LLM is used for future processing.

The invention relates to an industrial equipment maintenance intelligent question-answering system based on multi-agent collaboration. Wherein the input unit is used for receiving text, voice, image or equipment scanning and other multi-mode user input information and analyzing the information into structured problem information; the scheduling unit performs semantic understanding and problem classification on the structured problem information based on the fine-tuned cross-language pre-training language model and an incremental training mechanism; the processing unit calls a corresponding domain agent according to the classification result, and generates an intelligent question and answer processing result including predictive maintenance suggestions, structured reply content and semantic annotation information; and the fusion unit fuses the local knowledge base, the graph database and the networking retrieval information, performs multi-hop semantic reasoning on the intelligent question and answer processing result, and generates multi-modal reply information including text description, image screenshots, prediction curves and recommendation links. The system can support multi-language and multi-mode intelligent question answering and predictive maintenance in a complex industrial maintenance scene.

In a described embodiment, a multi-agent system for processing information is provided including a data processing agent configured to ingest and normalize raw data inputs to produce standardized data and a standards integration agent configured to apply reporting standards into the standardized data thereby generating integrated reporting standards. The system further includes a performance alignment agent configured to align performance indicators based on the standardized data and the integrated reporting standards and an information synthesis agent configured to process narrative information from the standardized data and the integrated reporting standards. An orchestration framework configured to manage operations of the data processing agent, the standards integration agent, and the performance alignment agent to produce a regulatory repot compliant with regulatory requirements is further provided. The orchestration framework is further executable by a large language model.

The invention discloses an automatic optimization method for cue words of a large language model, and relates to the technical field of cue word optimization. Comprises: setting an initial cue word; constructing training data, gradient generation cue word templates, cue word editing templates, optimizers and task models; randomly sampling a small batch of data from the training data as a training set, and predicting a sample of the training set by the task model according to the initial cue word to obtain an answer; comparing the answer with a sample label of a training set to obtain an error example set; inputting the error example set and the initial cue word into a gradient generation cue word template, and generating a gradient analysis request; the optimizer model receives and analyzes the gradient analysis request to generate a second natural language gradient; and inputting the second natural language gradient and the initial cue word into a cue word editing template, and modifying the initial cue word by the optimizer model according to the cue word editing template to generate a second candidate cue word. The technical problem that the efficiency of manually writing prompt words is low is solved.

Various embodiments of the technology described programmatically access a user query intended for a Large Language Model (LLM), analyze the user query, and determine prompt-enriching information that is combined with the user query to generate an enriched user query that is ultimately communicated to the LLM. In this manner, additional prompt-enriching information or context is added to the user query before being communicated to the LLM so that the additional prompt-enriching information, along with the user query, can be tokenized to better guide the LLM to a more accurate answer without modifying weights, parameters, or training of the LLM. Certain embodiments have the technical effect of improved accuracy relative to existing approaches by enriching user queries with prompt-enriching information to generate an enriched user query that is passed to the LLM. Based on the enriched user query, certain embodiments reduce the likelihood of hallucinations present in the LLM response.

Herein disclosed is receiving predetermined content, receiving a request to transform the predetermined content into a derivative work, receiving a requested theme for the derivative work, using generative artificial intelligence to create the derivative work generated as a function of the predetermined content and the requested theme, determining if the generated derivative work is approved, in response to determining the generated derivative work is approved, applying a digital watermark to the approved derivative work, configuring an authorization server to govern use of the approved derivative work based on the digital watermark and providing user access to the authorized derivative work. The requested theme may be determined using a Large Language Model (LLM) and a chatbot interview. The generative artificial intelligence may comprise a diffusion model. The content may comprise music.

The invention relates to the technical field of text data analysis, in particular to an electronic medical record free text analysis method, system and device, which improve the efficiency of information extraction and reduce the demand of manual intervention. The method comprises the steps of receiving free text data of the electronic medical record, performing cleaning, word segmentation and medical term standardization processing, and converting an unstructured text into structured data; based on deep learning and a medical knowledge base, text features are extracted through a pre-training language model, entity boundaries are captured on an output layer in combination with a conditional random field, medical entities in a text are recognized, and the recognized entities are classified and labeled; extracting a causal relationship, a treatment relationship and an examination relationship among entities through a dependency syntactic analysis and semantic role labeling technology, and constructing an entity association network; performing dynamic correction and supplementation on entity classification and relationships in combination with medical record context information; and outputting an analysis result in a structured JSON format to generate a computable semantic map.

The invention relates to the technical field of industrial equipment state monitoring and intelligent fault diagnosis, and discloses an industrial fault diagnosis method and system based on natural language fault features and large model knowledge enhanced reasoning, and the method comprises the steps: obtaining a multi-source monitoring and state analysis result from detected equipment in an equipment operation stage; converting the features into a fault feature text of a structured natural language; carrying out vectorization coding on the fault feature text, executing similarity retrieval in a pre-constructed industrial fault knowledge base, and recalling knowledge fragments; and based on the fault feature text and the recall knowledge fragment, constructing a reasoning prompt word, inputting the reasoning prompt word into a large language model for knowledge enhanced reasoning, and generating a diagnosis result. According to the method, the problems that numerical evidence and semantic knowledge are difficult to unify, the knowledge coverage and updating cost is high, and cross-working-condition migration and conclusion consistency are insufficient in an existing intelligent diagnosis technology based on a rule base or a knowledge graph are effectively solved.

A system for improving employee engagement and automating the integration of business processes. The system includes: a user interaction module configured to receive multimodal inputs, including natural language text and voice requests from employees via enterprise portals, mobile applications, voice-activated devices, and collaboration platforms, and to generate real-time responses via conversational output channels; a context processor that is operationally coupled with the user interaction module, wherein the context processor stores the interaction history, employee preferences, organizational role metadata, and task results in both short-term and long-term memory and dynamically derives context for controlling responses and initiating workflows; A natural language and intent processor that is communicatively linked to the context processor. The intent engine consists of large language models trained on company-specific lexicons to perform intent detection, entity extraction, ambiguity resolution, and sentiment or urgency classification from employee queries; a workflow orchestration layer in communication with the intent engine and the context processor, wherein the workflow orchestration layer includes a rule-based and AI-powered process execution engine configured to automatically initiate, route, and complete cross-functional business tasks, including approvals, escalations, and compliance checks, with workflows defined using modular templates that include conditional logic and time-based triggers; Webhooks and authentication protocols provide secure interoperability between the workflow orchestration layer and external enterprise platforms; a feedback and learning module that is operationally coupled with the workflow orchestration layer and the natural language understanding engine, wherein the feedback and learning module is configured to analyze task completion rates, response accuracy, latency metrics, and user feedback signals, and to retrain underlying language and workflow models for adaptive improvement in real time; and an administration console with role-based access controls, compliance dashboards, audit trails and interfaces for workflow configuration, whereby the administration console enables authorized personnel to monitor system operations, adjust interaction rules and enforce data protection restrictions across departments.

The invention discloses a large model scheduling multi-agent power grid fault coping strategy knowledge graph extension method and system, and the method comprises the steps: carrying out the modeling of task distribution as a mixed integer programming problem, and achieving the solving through a relaxation-correction algorithm; extracting and normalizing a core task of knowledge, and outputting structured data; summarizing the obtained text segments, and prompting a large language model to retain key entities, relationships and domain-specific terms; using an LLM-based named entity recognition technology, combining with prompt and dictionary / ontology filtering in the power dispatching field, recognizing related entities in a text, and normalizing the related entities into a standard form in a knowledge graph; detecting logic contradictions between the newly extracted triples and existing relationships in the knowledge graph, and classifying and solving the contradictions by utilizing debate prompts based on LLM (Logistics Library Model); summarizing the plurality of verification signals, and calculating the global confidence, definition and correlation score of each triple;