Artificial Intelligence / Machine Learning Model Drift Detection and Correction for Robotic Process Automation
The solution addresses AI/ML model drift in RPA by analyzing data and execution results to trigger retraining and human validation, ensuring accurate and reliable model performance in RPA systems.
Patent Information
- Application Number
- JP2023503130
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2020-07-21
- Filing Date
- 2020-09-24
- Publication Date
- 2024-11-11
- Estimated Expiration
- 2040-09-24
AI Technical Summary
AI/ML models in robotic process automation (RPA) can experience drift over time, leading to inaccurate predictions without user intervention, posing risks in mission-critical systems.
A computer program and method for detecting and correcting AI/ML model drift by analyzing input data and execution results, triggering retraining when change thresholds are met, and validating performance using statistical distributions and human review.
Ensures accurate and reliable AI/ML model performance by automatically detecting and correcting drift, enhancing the reliability of RPA systems.
Smart Images

Figure 0007580569000001 
Figure 0007580569000002 
Figure 0007580569000003
Abstract
Description
[Technical field]
[0001] (CROSS-REFERENCE TO RELATED APPLICATIONS) This application claims the benefit of U.S. Nonprovisional Patent Application No. 16 / 934,369, filed July 21, 2020. The subject matter of this previously filed application is hereby incorporated by reference in its entirety.
[0002] The present invention relates generally to robotic process automation (RPA), and more specifically to artificial intelligence (AI) / machine learning (ML) model drift detection and correction for RPA. [Background technology]
[0003] Trained AI / ML models may be deployed to provide probabilistic solutions for a variety of tasks, such as image recognition and document processing. However, predictions by AI / ML models may change or drift over time. AI / ML model drift may be particularly problematic in contexts where software robots may utilize AI / ML models to perform tasks, potentially with or without a user's knowledge or involvement at all. Improved techniques for detecting and / or correcting AI / ML model drift may therefore be beneficial. Summary of the Invention
[0004] Certain embodiments of the present invention may provide solutions to problems and needs in the field that have not yet been fully identified, appreciated, or solved by current RPA technology. For example, some embodiments of the present invention relate to detecting and correcting AI / ML model drift for RPA.
[0005] In an embodiment, a computer program for performing AI / ML model drift detection and correction for RPA is stored on a non-transitory computer-readable medium. The computer program is configured for at least one processor to analyze information related to input data for the AI / ML model to determine whether data drift has occurred, analyze information related to results from performance of the AI / ML model to determine whether model drift has occurred, or both. If, based on the analysis of the information, a change condition is found and / or a change threshold is met and / or exceeded, the computer program is further configured for the at least one processor to retrain the AI / ML model.
[0006] In an embodiment, a computer-implemented method for AI / ML model drift detection and correction for RPA includes analyzing information related to input data for an AI / ML model to determine whether data drift has occurred, analyzing information related to results from performance of the AI / ML model to determine whether model drift has occurred, or both. If a change condition is found based on the analysis of the information and a change threshold is met and / or exceeded, the method includes retraining the AI / ML model, verifying that the retrained AI / ML model meets one or more performance thresholds, and deploying the retrained AI / ML model if the retrained AI / ML model meets the one or more performance thresholds.
[0007] In yet another embodiment, a computer-implemented method for AI / ML model drift detection and correction for RPA includes analyzing information including one or more statistical distributions of predictions made by an AI / ML model. The information relates to input data of the AI / ML model to determine whether data drift has occurred, results from execution of the AI / ML model to determine whether model drift has occurred, or both. The method also includes mapping the one or more statistical distributions to respective actions taken by a plurality of RPA robots. If, based on the analysis of the information, a change condition is found and a change threshold is met and / or exceeded for at least a predetermined period of time, the method further includes retraining the AI / ML model. The action is performed at a respective activity of an RPA workflow of each of said RPA robots. [Brief description of the drawings]
[0008] So that the advantages of certain embodiments of the invention may be readily understood, a more particular description of the invention briefly described above will now be depicted with reference to specific embodiments which are illustrated in the accompanying drawings. It is to be understood that these drawings depict only typical embodiments of the invention and are therefore not to be considered as limiting its scope, but the invention will be depicted and explained with further specificity and detail through the use of the following accompanying drawings, in which:
[0009] [Figure 1] FIG. 1 is an architectural diagram illustrating a robotic process automation (RPA) system in accordance with an embodiment of the present invention.
[0010] [Diagram 2] FIG. 1 is an architectural diagram illustrating a deployed RPA system according to an embodiment of the present invention.
[0011] [Diagram 3]FIG. 2 is an architecture diagram illustrating the relationships between designers, activities, and drivers according to an embodiment of the present invention.
[0012] [Figure 4] FIG. 1 is an architectural diagram illustrating an RPA system according to an embodiment of the present invention.
[0013] [Diagram 5] FIG. 1 is an architectural diagram illustrating a computing system configured to perform AI / ML model drift detection and correction for RPA, according to an embodiment of the present invention.
[0014] [Figure 6] FIG. 1 is an architectural diagram illustrating a system configured to perform AI / ML model drift detection and correction for RPA, according to an embodiment of the present invention.
[0015] [Figure 7A] 1 is a flowchart illustrating a process for performing AI / ML model drift detection and correction for RPA in accordance with an embodiment of the present invention. [Figure 7B] 1 is a flowchart illustrating a process for performing AI / ML model drift detection and correction for RPA in accordance with an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0016] Unless otherwise noted, like reference characters denote corresponding features consistently throughout the attached drawings.
[0017] Detailed Description of the Embodiments Some embodiments relate to AI / ML model drift detection and correction for RPA. Some embodiments store key performance indicators (KPIs), such as how many predictions the AI / ML model made and how many were actually used, one or more statistical distributions of the predictions, etc. If a change threshold is met or exceeded, an alert or retraining trigger can be generated. For example, the change threshold can be a deviation from past performance for at least a certain amount, potentially for at least a certain period of time. The AI / ML model can then be trained using new (newer) data, potentially provided by one or more users and / or robots, and a new version of the AI / ML model or an entirely new model can be deployed to provide better predictions and resolve the drift issue.
[0018] AI / ML model drift can occur due to data drift, model drift, or both. Data drift is when the characteristics of the input data for an AI / ML model change over time from the data used to train the AI / ML model. This can lead to a decrease in the accuracy of predictions made by the AI / ML model. Model drift is a change over time in an underlying problem in the relationship between the input data to the AI / ML model and the output data from the AI / ML model. When either or both of these forms of drift occur, the AI / ML model may perform poorly or become inaccurate for the problem it was trained to solve. Data drift and model drift are collectively referred to herein as "concept drift."
[0019] In some embodiments, the statistical distributions are mapped to the actions that the RPA robot(s) are taking in its workflow(s). In certain embodiments, the statistical distributions may be matched to business outcomes. For example, consider a particular business outcome where a scale of 1-5 is used and outcomes typically fall within one standard distribution of the 1-5 scale (e.g., 2-4). If 1s and 5s start to become more common (e.g., occur more frequently than one standard deviation), an exception may be automatically triggered. A human reviewer may view the performance characteristics of the AI / ML model to determine how the AI / ML model has drifted. Human validation may be performed to improve the model and improve business outcomes. The root cause of drift may be either data drift or model drift. In some embodiments, a dashboard may be presented that displays these statistical distributions to the user for each business outcome.
[0020] In the context of RPA, drift detection can be especially important. For example, if an RPA robot is invoking an AI / ML model to make predictions, there is currently no existing way to check the quality of the model without human supervision in these scenarios. RPA robots are consumers of AI / ML models, and they may operate autonomously. Users may not have control over the actions the robots take based on the predictions of these AI / ML models, and such RPA robot autonomy may lead to outcomes that may be undesirable or even dangerous (e.g., in the case of mission-critical systems).
[0021] Some embodiments use one or more algorithms to measure data drift and model drift. For example, techniques to reduce the dimensionality of a data set (e.g., principal component analysis (PCA)), density estimation (i.e., evaluating the continuity of density in a high-dimensional space), combinations thereof, etc. may be used. However, it should be noted that any suitable technique(s) may be used to measure drift without departing from the scope of the present invention.
[0022] In certain embodiments, a framework is provided and enforced by the automation platform for how to detect and handle AI / ML model drift. For example, when an AI / ML model is deployed to production, it may have an input schema in common with other AI / ML models deployed to the automation platform. The degree to which the schema definition is enforced may depend on the AI / ML model itself and the AI / ML model pipeline in some embodiments. The input schema describes what the AI / ML model receives as input. In some embodiments, this may be a comma-separated value (CSV) file with predefined columns and specific data types. This may enforce specific field requirements and / or values.
[0023] For example, if an email body is passed as input to an AI / ML model in serialized JavaScript Object Notation (JSON), the email body may be treated as input. However, a CSV file containing numerical data may not be accepted because the AI / ML model may not know how to interpret the data. Thus, a schema can help prevent undesired or disallowed data formats from being fed to an AI / ML model.
[0024] As the deployed AI / ML model runs and analyses are performed, a representation of the input data and the AI / ML model predictions can be gradually constructed. For example, the distance between one-dimensional distributions of the original and new datasets can be measured for covariate drift. In some embodiments, the analysis includes determining whether statistical moments change. In certain embodiments, the statistical moments include mean, variance, skewness, kurtosis, covariance, or combinations thereof. Also, the distance between residual drift can be checked for the two models. For example, if two variations of an AI / ML model (e.g., a new version of the model and an earlier version of the model) are trained on different datasets or different parts of the dataset, the distance between the distributions of the models can be used to try to determine why drift is likely to occur in one model and not (or at least less likely to) occur in the other.
[0025] Consider again the example above, where the AI / ML model predicts five classes where the input data falls into a range of 1-5. The AI / ML model is deployed for six months, and 95% of the predictions are for classes 2, 3, or 4. However, the AI / ML model then begins to make more predictions for classes 1 and 5. It can then be determined whether the characteristics of the input data (e.g., distribution) have changed, causing a change in the predictions, or whether the change in the AI / ML model's predictions occurred without any significant change in the input data.
[0026] In some embodiments, activities in an RPA workflow may detect whether drift is occurring (and possibly what type). For example, a collection of activities may be available in an RPA designer application that allows developers to consider data or model drift. These activities may be configured based on drift occurring at a certain rate, exceeding a standard deviation, an uncommon prediction becoming significantly more common, etc.
[0027] If drift is determined to be occurring, a trigger may be generated. For example, the trigger may cause predictions made by the AI / ML model and / or the input data to be sent and stored for analysis by a human reviewer. An action center queue may allow a human to validate and automatically retrain the new version of the model. The human reviewer may then review what the AI / ML model predicted based on the input data and determine whether there are issues with the accuracy of the model. If the retrained AI / ML model meets a certain threshold (e.g., a confidence threshold, some kind of random sampling, other heuristics, etc.), it may be automatically deployed alongside the existing AI / ML model, and the drift of the model may be measured side-by-side to see if the new version of the AI / ML model is performing better on real data. If the model is below that threshold, a human review may be performed to determine why the threshold is not being met, and an attempt may be made to train the AI / ML model so that it meets or exceeds the threshold.
[0028] In some embodiments, when an AI / ML model is trained, a portion of the training data is split off for later use in testing. A collection of metrics can be used to create one high-level metric for evaluating the model. An example is the F1 score of a document understanding model, which is a derivative of precision and recall, and there is a trade-off between the two. Cross-validation is performed to split another portion of the data, mix it up, and test the model with the mixed data. However, if some of the split data is not representative of real-world data, the model may overfit the training data and not perform well in real-world applications.
[0029] In some embodiments, this process may be automated. If the F1 score is, for example, above 90%, a new version of the AI / ML model may be deployed. This new version may then be checked for drift and / or verified by a human.
[0030] Certain embodiments may be employed in Robotic Process Automation (RPA). FIG. 1 is an architectural diagram illustrating an RPA system 100, according to an embodiment of the present invention. The RPA system 100 includes a designer 110 that enables developers to design and implement workflows. The designer 110 provides solutions for application integration, as well as automating third-party applications, management information technology (IT) tasks, and business IT processes. The designer 110 may facilitate the development of automation projects, which are graphical representations of business processes. In simple terms, the designer 110 facilitates the development and deployment of workflows and robots.
[0031] Automation projects enable rule-based process automation by giving developers control over the order of execution and relationships between custom sets of steps developed in workflows, defined herein as "activities." One commercial example of an embodiment of the Designer 110 is UiPath Studio™. Each activity may include an action such as clicking a button, reading a file, writing to a log panel, etc. In some embodiments, workflows may be nested or embedded.
[0032] The types of workflows may include, but are not limited to, sequences, flowcharts, FSMs, and / or global exception handlers. Sequences may be particularly suitable for linear processes, allowing flow from one activity to another without cluttering the workflow. Flowcharts may be particularly suitable for more complex business logic, allowing for the integration of decisions and the connection of activities in more diverse ways through multiple branching logic operators. FSMs may be particularly suitable for large workflows. FSMs may use a finite number of states triggered by conditions (i.e., transitions) or activities during their execution. Global exception handlers may be particularly suitable for determining the behavior of a workflow when an execution error is encountered, or for debugging the process.
[0033] Once a workflow is developed in Designer 110, the execution of the business process is orchestrated by Conductor 120, which coordinates one or more Robots 130 that execute the workflow developed in Designer 110. One commercial example of an embodiment of Conductor 120 is UiPath Orchestrator™. Conductor 120 facilitates management of the creation, monitoring, and deployment of resources in an environment. Conductor 120 may act as, or one of, the integration points with third-party solutions and applications.
[0034] The conductor 120 may manage all the robots 130, connecting and executing them from a central point. Types of robots 130 that may be managed include, but are not limited to, attended robots 132, unattended robots 134, development robots (similar to unattended robots 134 but used for development and testing purposes), and non-production robots (similar to attended robots 132 but used for development and testing purposes). Attended robots 132 are triggered by user events and operate side-by-side with humans on the same computing system. Attended robots 132 may be used with the conductor 120 for a centralized process deployment and logging medium. Attended robots 132 may assist human users in accomplishing various tasks and may be triggered by user events. In some embodiments, processes cannot be initiated from the conductor 120 on this type of robot and / or they cannot be run under a locked screen. In certain embodiments, the attended robot 132 may only be launched from the robot tray or from a command prompt. The attended robot 132 preferably operates under human supervision in some embodiments.
[0035] Unattended robots 134 operate unattended in a virtual environment and can automate many processes. Unattended robots 134 can be responsible for providing remote execution, monitoring, scheduling, and work queue support. Debugging for all robot types can be performed in the designer 110 in some embodiments. Both attended and unattended robots can automate a variety of systems and applications, including, but not limited to, mainframes, web applications, VMs, enterprise applications (e.g., those produced by SAP®, SalesForce®, Oracle®, etc.), and computing system applications (e.g., desktop and laptop applications, mobile device applications, wearable computer applications, etc.).
[0036] The conductor 120 may have various capabilities including, but not limited to, provisioning, deployment, versioning, configuration, queuing, monitoring, logging, and / or providing interconnectivity. Provisioning may include creating and maintaining connections between the robots 130 and the conductor 120 (e.g., web applications). Deployment may include ensuring correct delivery of package versions to the robots 130 assigned for execution. Versioning may include managing unique instances of some processes or configurations in some embodiments. Configuration may include maintaining and delivering robot environments and process configurations. Queuing may include providing management of queues and queue items. Monitoring may include tracking robot specific data and maintaining user permissions. Logging may include storing and indexing logs in a database (e.g., an SQL database) and / or another storage mechanism (e.g., ElasticSearch®, which provides the ability to store large data sets and perform queries quickly). The conductor 120 may provide interconnectivity by operating as a central point of communication for third-party solutions and / or applications.
[0037] The robot 130 is an execution agent that executes the workflow built in the designer 110. One commercial example of some embodiments of the robot(s) 130 is UiPath Robots™. In some embodiments, the robot 130 installs the Microsoft Windows Service Control Manager (SCM) management service by default. As a result, such a robot 130 can open an interactive Windows session under the local system account and may have Windows service rights.
[0038] In some embodiments, the robot 130 may be installed in user mode, meaning that for such a robot 130, it has the same rights as the user to which the given robot 130 is installed. This feature may also be available for high density (HD) robots, ensuring maximum utilization of each machine. In some embodiments, either type of robot 130 may be configured in an HD environment.
[0039] The robot 130 in some embodiments is divided into multiple components, each specialized for a particular automation task. The robot components in some embodiments include, but are not limited to, the SCM managed robot service, the user mode robot service, the executor, the agent, and the command line. The SCM managed robot service manages and monitors the Windows session and acts as a proxy between the conductor 120 and the execution host (i.e., the computing system on which the robot 130 executes). These services are entrusted with and manage the credentials of the robot 130. The console application is launched by the SCM under the local system.
[0040] The user-mode robot service in some embodiments manages and monitors the Windows session and acts as a proxy between the conductor 120 and the execution host. The user-mode robot service may be delegated and manage the credentials of the robot 130. If the SCM management robot service is not installed, a Windows application may be launched automatically.
[0041] An Executor may execute a given job under a Windows session (i.e., execute a workflow). An Executor may be aware of per-monitor dots-per-inch (DPI) settings. An Agent may be a Windows Presentation Foundation (WPF) application that displays available jobs in a system tray window. An Agent may be a client of a service. An Agent may ask to start or stop a job or change settings. A Command Line is a client of a service. A Command Line is a console application that can request the start of a job and wait for its output.
[0042] As described above, the separation of the components of the robot 130 helps developers, support users, and computing systems to more easily execute, identify, and track what each component is doing. In this way, special behaviors can be configured for each component, such as setting different firewall rules for executors and services. The executor may always be aware of per-monitor DPI settings in some embodiments. As a result, workflows may execute at any DPI regardless of the configuration of the computing system on which the workflow was created. Also, in some embodiments, projects from the designer 110 may be made independent of the browser zoom level. For applications that are not DPI aware or are marked as not DPI aware intentionally, in some embodiments, DPI may be disabled.
[0043] FIG. 2 is an architecture diagram illustrating a deployed RPA system 200 according to an embodiment of the present invention. In some embodiments, the RPA system 200 may be or be part of the RPA system 100 of FIG. 1. It should be noted that the client side, the server side, or both may include any desired number of computing systems without departing from the scope of the present invention. On the client side, the robot application 210 includes an executor 212, an agent 214, and a designer 216. However, in some embodiments, the designer 216 may not be running on the computing system 210. The executor 212 is running a process. As shown in FIG. 2, multiple business projects may be running simultaneously. The agent 214 (e.g., a Windows service) is a single connection point for all the executors 212 in this embodiment. All messages in this embodiment are logged to the conductor 230, which further processes them via the database server 240, the indexer server 250, or both. As described above with respect to FIG. 1, the executor 212 may be a robotic component.
[0044] In some embodiments, a robot represents an association between a machine name and a username. A robot may manage multiple executors simultaneously. In computing systems that support multiple interactive sessions running simultaneously (such as Windows Server 2012), multiple robots may run simultaneously, each running in a separate Windows session using a unique username. This is referred to as the HD Robot above.
[0045] The agent 214 is also responsible for transmitting the robot's status (e.g., periodically sending "heartbeat" messages to indicate that the robot is still functioning) and downloading required versions of packages to be fulfilled. Communication between the agent 214 and the conductor 230 is, in some embodiments, always initiated by the agent 214. In a notification scenario, the agent 214 may open a WebSocket channel that is later used by the conductor 230 to send commands (e.g., start, stop, etc.) to the robot.
[0046] The server side includes a presentation layer (web application 232, Open Data Protocol (OData) Representational State Transfer (REST) Application Programming Interface (API) endpoints 234, notifications and monitoring 236), a service layer (API implementation / business logic 238), and a persistence layer (database server 240, indexer server 250). Conductor 230 includes web application 232, OData REST API endpoints 234, notifications and monitoring 236, and API implementation / business logic 238. In some embodiments, most actions that a user performs in the conductor 230 interface (e.g., via browser 220) are performed by calling various APIs. Such operations may include, but are not limited to, launching jobs on a robot, adding / removing data in a queue, scheduling jobs to run unattended, etc., without departing from the scope of the invention. Web application 232 is the visual layer of the server platform. In this embodiment, web application 232 uses Hypertext Markup Language (HTML) and JavaScript (JS). However, any desired markup language, scripting language, or any other format may be used without departing from the scope of the present invention. A user interacts with web pages from web application 232, in this embodiment via browser 220, to perform various operations to control conductor 230. For example, a user may create robot groups, assign packages to robots, analyze per-robot and / or per-process logs, start and stop robots, etc.
[0047] In addition to the web application 232, the conductor 230 also includes a services layer that exposes an OData REST API endpoint 234. However, other endpoints may be included without departing from the scope of the invention. The REST API is consumed by both the web application 232 and the agent 214, which in this embodiment is a supervisor of one or more robots on a client computer.
[0048] The REST API of the present embodiment covers configuration, logging, monitoring, and queuing functionality. The configuration endpoint may be used in some embodiments to define and configure users, permissions, robots, assets, releases, and environments of an application. The logging REST endpoint may be used to log various information, such as errors, explicit messages sent by the robot, and other environment specific information. The deployment REST endpoint may be used by the robot to query the version of the package that should be executed when a start job command is used in the conductor 230. The queuing REST endpoint may be responsible for managing queues and queue items, such as adding data to the queue, getting transactions from the queue, and setting the status of transactions.
[0049] Monitoring REST endpoints may monitor the web application 232 and the agents 214. Notification and monitoring API 236 may be a REST endpoint used to register agents 214, deliver configuration settings to agents 214, and send and receive notifications from the server and agents 214. Notification and monitoring API 236 may use WebSocket communications in some embodiments.
[0050] The persistence layer, in this embodiment, includes a pair of servers - a database server 240 (e.g., SQL Server) and an indexer server 250. The database server 240 in this embodiment stores configurations for robots, robot groups, associated processes, users, roles, schedules, etc. This information is managed, in some embodiments, via a web application 232. The database server 240 may also manage queues and queue items. In some embodiments, the database server 240 may also store messages logged by robots (in addition to or instead of the indexer server 250).
[0051] Optionally in some embodiments, indexer server 250 stores and indexes information logged by the robots. In certain embodiments, indexer server 250 may be disabled via a configuration setting. In some embodiments, indexer server 250 uses ElasticSearch®, a full-text search engine from an open source project. Messages logged by the robots (e.g., using activities such as log messages or line writes) may be sent via logging REST endpoint(s) to indexer server 250, where they are indexed for future use.
[0052] FIG. 3 is an architecture diagram illustrating the relationship 300 between a designer 310, activities 320, 330, and a driver 340, according to an embodiment of the present invention. As can be seen, a developer uses the designer 310 to develop a workflow to be performed by the robot. The workflow may include user-defined activities 320 and UI automation activities 330. In some embodiments, non-textual visual components in an image may be identified, referred to herein as computer vision (CV). Some CV activities related to such components may include, but are not limited to, click, type, get text, hover, detect presence or absence of element, update scope, highlight, and the like. In some embodiments, click may identify an element and click on it, for example, using CV, optical character recognition (OCR), fuzzy text matching, and multi-anchors. Type may identify an element using the above and type within an element. Get text may locate a particular text and scan it using OCR. Hover may identify an element and hover over it. Detect presence or absence of element may check to detect the presence or absence of an element on the screen using the techniques described above. In some embodiments, there may be hundreds or even thousands of activities that may be implemented in the designer 310. However, any number and / or types of activities may be utilized without departing from the scope of the present invention.
[0053] UI automation activities 330 are a subset of specialized low-level activities that are written in low-level code (e.g., CV activities) and facilitate interactions with the screen. UI automation activities 330 facilitate these interactions through drivers 340 that enable the robot to interact with desired software. For example, drivers 340 may include OS drivers 342, browser drivers 344, VM drivers 346, enterprise application drivers 348, etc.
[0054] Drivers 340 may interact with the OS at a low level, looking for hooks, monitoring keys, etc. They may facilitate integration with Chrome®, IE®, Citrix®, SAP®, etc. For example, a "click" activity plays the same role in these different applications via drivers 340.
[0055] FIG. 4 is an architecture diagram illustrating an RPA system 400, according to an embodiment of the present invention. In some embodiments, the RPA system 400 may be or may include the RPA systems 100 and / or 200 of FIG. 1 and / or FIG. 2. The RPA system 400 includes multiple client computing systems 410 that execute robots. The computing systems 410 may communicate with a conductor computing system 420 via web applications executed thereon. The conductor computing system 420 may in turn communicate with a database server 430 and an optional indexer server 440.
[0056] 1 and 3, it should be noted that although web applications are used in these embodiments, any suitable client / server software may be used without departing from the scope of the present invention. For example, the conductor may run a server-side application on a client computing system that communicates with a non-web-based client software application.
[0057] FIG. 5 is an architecture diagram illustrating a computing system 500 configured to perform AI / ML model drift detection and correction for RPA, according to an embodiment of the present invention. In some embodiments, the computing system 500 may be one or more computing systems depicted and / or described herein. The computing system 500 includes a bus 505 or other communication mechanism for communicating information and a processor(s) 510 coupled to the bus 505 for processing information. The processor(s) 510 may be any type of general or application-specific processor, including a central processing unit (CPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a graphics processing unit (GPU), multiple instances thereof, and / or any combination thereof. The processor(s) 510 may also have multiple processing cores, and at least some of the cores may be configured to perform specific functions. In some embodiments, multiple parallel processing may be used. In certain embodiments, at least one of the processor(s) 510 may be a neuromorphic circuit that includes processing elements that mimic biological neurons. In some embodiments, neuromorphic circuitry may not require typical components of a von Neumann computing architecture.
[0058] The computing system 500 further includes a memory 515 for storing information and instructions executed by the processor(s) 510. The memory 515 may be comprised of random access memory (RAM), read-only memory (ROM), flash memory, cache, static storage such as a magnetic or optical disk, or other types of non-transitory computer-readable media, or any combination thereof. The non-transitory computer-readable media may be any available media accessible by the processor(s) 510 and may include volatile media, non-volatile media, or both. Additionally, the media may be removable, non-removable, or both.
[0059] Additionally, the computing system 500 includes a communication device 520, such as a transceiver, to provide access to a communication network via wireless and / or wired connections. In some embodiments, the communications device 520 may be configured to support any of the following wireless technologies: Frequency Division Multiple Access (FDMA), Single Carrier FDMA (SC-FDMA), Time Division Multiple Access (TDMA), Code Division Multiple Access (CDMA), Orthogonal Frequency Division Multiple Access (OFDM), Orthogonal Frequency Division Multiple Access (OFDMA), Global System for Mobile (GSM) communications, General Packet Radio Service (GPRS), Universal Mobile Telecommunications System (UMTS), cdma2000, Wideband CDMA (W-CDMA), High-Speed Downlink Packet Access (HSDPA), High-Speed Uplink Packet Access (HSUPA), High-Speed Packet Access (HSPA), Long Term Evolution (LTE), LTE Advanced (LTE-A), LTE-Advanced (LTE-B), LTE-Advanced-B, LTE-Advanced-C ... Advanced), 802.11x, Wi-Fi, Zigbee, Ultra-Wide Band (UWB), 802.16x, 802.15, Home Node-B (HnB), Bluetooth, Radio Frequency Identification (RFID), Infrared Data Association (IrDA), Near-Field Communications (NFC), 5th Generation (5G), New Radio (NR), any combination thereof, and / or any other currently existing or future implemented communications standards and / or protocols without departing from the scope of the present invention.In some embodiments, the communications device 520 may include one or more antennas that are a singular antenna, an array of antennas, phased antennas, switched antennas, beamforming antennas, beamsteering antennas, combinations thereof, and / or any other antenna configuration without departing from the scope of the invention.
[0060] The processor(s) 510 are further coupled via bus 505 to a display 525, such as a plasma display, a liquid crystal display (LCD), a light emitting diode (LED) display, a field emission display (FED), an organic light emitting diode (OLED) display, a flexible OLED display, a flexible substrate display, a projection display, a 4K display, a high definition display, a Retina® display, an in-plane switching (IPS) display, or any other suitable display for displaying information to a user. The display 525 may be configured as a touch (haptic) display, a three-dimensional (3D) touch display, a multi-input touch display, a multi-touch display, or the like, using resistive, capacitive, surface acoustic wave (SAW) capacitive, infrared, optical imaging, dispersive signaling, acoustic pulse recognition, frustrated total internal reflection, or the like. Any suitable display device and haptic I / O may be used without departing from the scope of the invention.
[0061] A keyboard 530 and cursor control device 535, such as a computer mouse, touchpad, and the like, are further coupled to the bus 505 to allow a user to interface with the computing system 500. However, in certain embodiments, a physical keyboard and mouse may not be present and the user may interact with the device solely through the display 525 and / or touchpad (not shown). Any type and combination of input devices may be used as a matter of design choice. In certain embodiments, no physical input devices and / or displays are present. For example, a user may interact with the computing system 500 remotely via another computing system in communication with it, or the computing system 500 may operate autonomously.
[0062] The memory 515 stores software modules that provide functionality when executed by the processor(s) 510. The modules include an operating system 540 for the computing system 500. The modules further include an AI / ML model drift detection and correction module 545 configured to perform all or a portion of the processes described herein, or derivatives thereof. The computing system 500 may include one or more additional functional modules 550 that include additional functionality.
[0063] Those skilled in the art will appreciate that the "system" may be embodied as a server, an embedded computing system, a personal computer, a console, a personal digital assistant (PDA), a mobile phone, a tablet computing device, a quantum computing system, or any other suitable computing device or combination of devices without departing from the scope of the present invention. Presenting the functions described above as being performed by a "system" is not intended to limit the scope of the present invention in any way, but rather to provide an example of many embodiments of the present invention. Indeed, the methods, systems, and apparatus disclosed herein may be implemented in localized and distributed forms consistent with computing techniques, including cloud computing systems.
[0064] It should be noted that some of the system features described herein are presented as modules to better emphasize implementation independence. For example, a module may be implemented as a hardware circuit comprising custom very large scale integrated (VLSI) circuits or gate arrays, off-the-shelf semiconductors such as logic chips, transistors, or other discrete components. A module may also be implemented in programmable hardware devices such as field programmable gate arrays, programmable array logic, programmable logic devices, graphics processing units, and the like.
[0065] Modules may also be implemented at least in part in software for execution by various types of processors. For example, an identified unit of executable code may include one or more physical or logical blocks of computer instructions, which may be organized, for example, as an object, procedure, or function. Nevertheless, executable identified modules need not be physically located together, but may include separate instructions stored in different locations, which when logically combined, comprise the modules, to accomplish the purpose stated for the modules. Furthermore, modules may be stored on non-transitory computer readable media, such as, for example, hard disk drives, flash devices, RAM, tape, and / or any other non-transitory computer readable media used to store data without departing from the scope of the present invention.
[0066] Indeed, a module of executable code may be a single instruction, or many instructions, and may even be distributed across several different code segments, different programs, and across several memory devices. Similarly, operational data may be identified and illustrated herein within a module, and may be embodied and organized in any suitable form within any suitable type of data structure. The operational data may be collected as a single data set, or may be distributed in different locations across different storage devices, or may exist, at least in part, simply as electronic signals on a system or network.
[0067] In some embodiments, information regarding the performance of the AI / ML model is obtained. This information may come from a single computing system for AI / ML models that are local to that user or computing system, from multiple or multiple computing systems for AI / ML models that are intended to be applied more generally (e.g., providing solutions at a group level, company level, global level, etc.), from the AI / ML model itself, from an application or other process that monitors the AI / ML model, any combination thereof, etc. This information may help determine how well the AI / ML model is performing over time. For example, the information may include, but is not limited to, what predictions were made (e.g., output from the AI / ML model), how many predictions the AI / ML model made and how many were actually used, one or more statistical distributions of the predictions (e.g., normal distribution, binomial distribution, or any other statistical distribution), the input data provided to the AI / ML model, any combination of these, etc. In some embodiments, the statistical distributions are mapped to actions taken by the RPA robot in its workflow (e.g., mapped to activities).
[0068] After this information is collected, it is analyzed to determine whether data drift, model drift, or both have occurred. When a change condition is found (e.g., predicted values or input data fall outside of historical ranges) and / or a change threshold is met or exceeded (e.g., the statistical performance of the model deviates from historical performance by at least a certain amount), an alert or retraining trigger is generated. In some embodiments, the change condition or change threshold must be met for at least a certain period or number of instances before the AI / ML model performance is considered to have drifted. In certain embodiments, the distance between the one-dimensional distributions of the original and new datasets may be measured for covariate drift. Also, the distance between residual drift may be checked in a version of the AI / ML model, if one already exists.
[0069] The AI / ML model may then be retrained using the collected information and / or other information to attempt to improve the AI / ML model performance. In some embodiments, if the retrained AI / ML model meets certain thresholds, it may be automatically deployed alongside the existing AI / ML model, and model drift may be measured side-by-side to see if the new version of the AI / ML model performs better on real data. In certain embodiments, the new version of the AI / ML model is not deployed unless it meets these thresholds.
[0070] In some embodiments, canary testing is used. For example, a new version of an AI / ML model may be used by a relatively small number of users (e.g., 10% of users at a time, by geography or some other filter) who may not be aware of the changes in the AI / ML model. If the new version of the AI / ML model does not perform as well as the previous version, the new version may be rolled back and retraining may be performed.
[0071] In certain embodiments, a champion / challenger technique may be used, where input data is sent to both the new version of the AI / ML model and the previous version of the AI / ML model, but only one set of outputs is used. This may help analyze how the new version of the AI / ML model and the previous version of the AI / ML model perform under apples-to-apples conditions based on some ground truth. This technique may help determine which AI / ML model version is appropriate or closer to being appropriate more frequently. This technique may also provide information useful for troubleshooting as to why the previous version of the AI / ML model exhibited a particular behavior.
[0072] In some embodiments, a common framework is provided and enforced by the automation platform regarding what information is fed to AI / ML models invoked by RPA robots from workflow activities and / or executed by RPA robots within workflow activities. This framework can help ensure that AI / ML models used by the RPA platform consume the intended data. This framework can also help ensure that AI / ML models receive inputs in the proper format to enforce specific field requirements and / or values.
[0073] FIG. 6 is an architecture diagram illustrating a system 600 configured to perform AI / ML model drift detection and correction for RPA, according to an embodiment of the present invention. The system 600 includes user computing systems, such as desktop computers 602, 603, tablets 604, smartphones 606, etc. However, any desired computing system may be used without departing from the scope of the present invention, including, but not limited to, smart watches, laptop computers, Internet of Things (IoT) devices, vehicle computing systems, etc. Also, while three user computing systems are shown in FIG. 6, any suitable number of computing systems may be used without departing from the scope of the present invention. For example, in some embodiments, tens, hundreds, thousands, or millions of computing systems may be used.
[0074] Each computing system 602, 604, 606 has an RPA robot 610 running thereon that invokes and / or executes an ML model (e.g., one of the AI / ML models 632 on the server 630, a local ML model on the computing system 602, 604, 606, an ML model located anywhere else, etc.). The RPA robot 610, in some embodiments, may be a robot created via an RPA designer application. In this embodiment, the RPA robot 610 sends requests, and potentially input data, to the AI / ML model 632. However, in some embodiments, the AI / ML model 632 may obtain input data from one or more other sources (e.g., a database 640, another computing system, etc.) in addition to or instead of the input data from the RPA robot 610.
[0075] The RPA robot 610 sends AI / ML model requests to and receives AI / ML model execution results from the AI / ML model 632 of the server 630 via a network 620 (e.g., a local area network (LAN), a mobile communication network, a satellite communication network, the Internet, any combination thereof, etc.). The server 630 collects and stores information related to model drift for the AI / ML model 632 in a database 640. This information may relate to data drift, model drift, or both.
[0076] If a change condition is found (e.g., predictions or input data fall outside historical ranges) and / or a change threshold is met or exceeded (e.g., model statistical performance deviates from historical performance by at least a certain amount) for one of the AI / ML models 632, in some embodiments the server 630 sends an alert or retraining trigger (e.g., as part of a queue for review and labeling) to an application 652 running on a reviewer computing system 650. In some embodiments, the application 652 may enable the reviewer to view information about the AI / ML model drift, revise the results (e.g., for model drift), and / or recognize that the input data has changed in nature (e.g., data drift). In some embodiments, the application 652 may enable the ML operations personnel who "own" the ML model in production to view the model drift in a two-dimensional graph. Dimensionality reduction may be performed, for example using PCA, to visually indicate certain anomalies. Specific alerts may also be created to flag when one of these anomalies occurs.
[0077] In some embodiments, there may be more layers of abstraction in application 652 where data is presented to business users in a way that allows them to understand the meaning of the data (e.g., 2D graphs showing business users how analysis of their business data has changed). This may be presented in addition to or instead of displays aimed at more technically savvy ML operations personnel. For example, a mechanism like a word cloud may be used to show more frequently occurring results in a larger font, making it easier for humans to see which results are most widely accepted.
[0078] In certain embodiments, significant changes may be highlighted. For example, a business user may be able to see that an AI / ML model for approving loans jumped from 5 loans approved in a day to 20 loans approved in a day. The business user may then define rules to modify the model's behavior, all the way until retraining occurs if the issue persists. The business user may also be able to determine if there were anomalies not due to the performance of the AI / ML model (e.g., the number of loan applications increased, more qualified applicants are applying, etc.). Humans may also use application 652 to determine for themselves what changes are causing deviations that were not detected by the automated drift detection mechanism. In some embodiments, there may be two layers, one for technical users and another for business users.
[0079] In particular embodiments, multiple or numerous computing systems and review applications, including computing system 650 and application 652, may be used to collectively review and modify the results. In some embodiments, these results are pulled from a common queue. These modifications may then be provided back to server 630 and stored in database 640 to retrain the drifting AI / ML model 632. However, in particular embodiments, server 630 may automatically detect that AI / ML model drift is occurring due to changes in input data and / or changes in AI / ML model predictions.
[0080] After AI / ML data and / or model drift is detected by server 630 (and in some embodiments after human correction), drift AI / ML model 632 is retrained. In some embodiments, the retraining is performed by server 630. However, in particular embodiments, the retraining may be performed by one or more other servers in addition to or instead of server 630. Drift AI / ML model 632 may then be retrained using the collected information and / or other information to attempt to improve the AI / ML model performance.
[0081] After the new version of the AI / ML model 632 is properly trained (e.g., if the retrained AI / ML model 632 meets a certain threshold), the server 630 may save the new version of the AI / ML model 632 to the database 640 and replace the previous version of the AI / ML model. However, in certain embodiments, multiple versions of the AI / ML model 632 may be stored. The server 630 then deploys the retrained AI / ML model 632 to be invoked by the RPA robot 610. However, in some embodiments, the retrained AI / ML model 632 is deployed alongside one or more previous versions of the AI / ML model 632. The drift of each version of the AI / ML model 632 is then measured side-by-side to see if the new version of the AI / ML model 632 is performing better on real data. If this is determined to be the case over time, the new version of the AI / ML model 632 may replace the previous version of the AI / ML model 632. However, if this is not the case, the new version of the AI / ML model may be discarded or further retraining may be performed.
[0082] 7A and 7B are flow charts illustrating a process 700 for performing AI / ML model drift detection and correction for RPA, according to an embodiment of the present invention. The process begins with receiving a request to execute an AI / ML model from an RPA robot at 705 and input information for the AI / ML model. Validating that the input information fits into the framework of the AI / ML model is performed at 710. If the input information does not fit into the framework at 715, an error is provided to the RPA robot indicating that the input information does not fit into the framework at 720. If the input information fits into the framework at 715, the AI / ML model is executed using the input information and the results of the execution of the AI / ML model are returned to the RPA robot at 725. In some embodiments, the framework is enforced for multiple AI / ML models.
[0083] Information about the input data of the AI / ML model to determine whether data drift has occurred, information about the results of the execution of the AI / ML model to determine whether model drift has occurred, or both are analyzed at 730. In some embodiments, the analysis includes measuring one or more distances between one-dimensional distributions of the original data set and the new data set for covariate drift. In certain embodiments, the analysis includes determining whether a statistical moment changes. In some embodiments, the statistical moment includes mean, variance, skewness, kurtosis, covariance, or a combination thereof. In certain embodiments, the information includes what predictions the AI / ML model made, how many predictions the AI / ML model made and how many predictions were used by the RPA robot, one or more statistical distributions of the predictions made by the AI / ML model, the input data provided to the AI / ML model, or a combination thereof. In some embodiments, the analysis of the information is performed across multiple RPA processes performed by multiple respective RPA robots (e.g., via federated learning).
[0084] In some embodiments, the information includes one or more statistical distributions of the predictions made by the AI / ML model. In such embodiments, the one or more statistical distributions may be mapped to respective actions taken by the multiple RPA robots at 735. In some embodiments, the actions are taken at respective activities of the RPA workflow of the respective RPA robots.
[0085] A dimensionality reduction technique is performed on the information to generate a two-dimensional representation of the information, and the two-dimensional representation of the information is displayed at 740. Based on the analysis of the information at 745, a change condition is found and a verification is performed whether a change threshold is met or exceeded, or both. In some embodiments, the verification includes verifying whether the change condition, the change threshold, or both, have been met or exceeded for at least a predetermined period of time.
[0086] Based on the analysis of the information at 750, if a change condition is found and the change threshold is met and / or exceeded, the AI / ML model is retrained at 755. The retrained AI / ML model is then verified at 760 to meet one or more performance thresholds. If the one or more performance thresholds are not met at 765, the process returns to retraining the AI / ML model at 755. However, if the retrained AI / ML model meets one or more performance thresholds at 765, the retrained AI / ML model is deployed at 770. In some embodiments, deploying the AI / ML model may include modifying one or more activities of one or more RPA robots to invoke the new AI / ML model, replacing a previous version of the AI / ML model with the retrained AI / ML model. In certain embodiments, the AI / ML model is deployed using a canary testing technique or a champion / challenger testing technique.
[0087] The model drift of the retrained AI / ML model and the previous version AI / ML model is measured at 775 using real data. If the retrained AI / ML model does not perform better on real data than the previous version AI / ML model based on the model drift measured at 780, the retrained AI / ML model is discarded at 785 and / or the process returns to step 755 for further retraining. However, if the retrained AI / ML model performs better on real data than the previous version AI / ML model based on the model drift measured at 780, the retrained AI / ML model is used in addition to or instead of the previous version AI / ML model at 790. Based on the input data and / or use case, the retrained AI / ML model or the previous version AI / ML model is selected and recalled at 795. In some embodiments, the selection is made in a hierarchical manner.
[0088] In some embodiments, model drift detection may be performed per process or per group of processes. For example, it may be good to use one version of an AI / ML model for one type of processing. It may be observed that when an AI / ML model is trained on input data with certain characteristics, the AI / ML model becomes less accurate for another type of input data. For example, consider an AI / ML model used to process invoices that was trained using a European date format (e.g., 17 November 2019 or 17 / 11 / 2019). If more invoices with a US date format are used during training (e.g., November 17, 2019 or 11 / 17 / 2019), the AI / ML model may start to fail for invoices with a European date format.
[0089] Thus, in some embodiments, different versions of the AI / ML model may be trained to respond to different situations and drift detection may be used to select the AI / ML model version that performs best for the given data. In certain embodiments, the AI / ML model versions may be applied in a hierarchical manner based on drift detection at the RPA workflow level. Logic may be developed to determine which AI / ML model version to use based on the type of input, or multiple AI / ML models may be tried in a hierarchical manner and the result selected if certain criteria are met. This may also provide insight into what drift is occurring in multiple AI / ML model versions. This may help determine if an AI / ML model is bad for one use case but not another. In certain embodiments, the AI / ML model chain process may be constructed such that a deterministic logic governs the AI / ML model version used in the hierarchy. In some embodiments, another AI / ML model may determine which AI / ML model version to use. In certain embodiments, federated learning may be used, where multiple computing systems collaboratively learn a shared predictive model while keeping the training data on the local device. This can be particularly beneficial for ensuring compliance with RPA governance requirements (e.g., the EU General Data Protection Regulation (GDPR), the US Health Insurance Portability and Accountability Act (HIPAA), third-party terms of use, etc.).
[0090] The process steps performed in Figures 7A and 7B may be performed by a computer program encoding instructions to a processor(s) to perform at least a portion of the process(es) described in Figures 7A and 7B according to an embodiment of the present invention. The computer program may be stored on a non-transitory computer-readable medium. The computer-readable medium may be, but is not limited to, a hard disk drive, a flash device, a RAM, a tape, and / or any other such medium or combination of media used to store data. The computer program may include coded instructions for controlling a processor(s) of a computing system (e.g., processor(s) 510 of computing system 500 of Figure 5) to implement all or a portion of the process steps described in Figures 7A and 7B, which may also be stored on a computer-readable medium.
[0091] The computer program may be implemented in hardware, software, or a hybrid implementation. The computer program may be composed of modules in operable communication with each other and designed to send information or instructions to a display. The computer program may be configured to run on a general purpose computer, an ASIC, or any other suitable device.
[0092] It will be readily understood that the components of the various embodiments of the present invention, as generally described and illustrated herein, may be arranged and designed in a wide variety of different configurations. Thus, the detailed description of the embodiments of the present invention, as represented in the accompanying figures, is not intended to limit the scope of the invention as claimed, but is merely representative of selected embodiments of the invention.
[0093] The features, structures, or characteristics of the invention described throughout this specification may be combined in any suitable manner in one or more embodiments. For example, reference throughout this specification to "certain embodiments," "some embodiments," or similar language means that the particular feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment of the invention. Thus, the appearance of "certain embodiments," "some embodiments," "other embodiments," or similar language throughout this specification does not necessarily refer to the same group of all embodiments, and the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
[0094] It should be noted that references to features, advantages, or similar language throughout this specification do not imply that all of the features and advantages that may be realized in the present invention should be in any single embodiment of the present invention, or that any embodiment of the present invention. Rather, language referring to features and advantages is understood to mean that the particular feature, advantage, or characteristic described in connection with an embodiment is included in at least one embodiment of the present invention. Thus, discussions of features and advantages throughout this specification, as well as similar language, may, but do not necessarily, refer to the same embodiment.
[0095] Furthermore, the described features, advantages, and characteristics of the invention may be combined in any suitable manner in one or more embodiments. Those skilled in the relevant art will recognize that the invention may be practiced without the specific features or advantages of one or more particular embodiments. In other instances, additional features and advantages may be recognized in certain embodiments, although they may not be present in all embodiments of the invention.
[0096] Those of ordinary skill in the art will readily appreciate that the invention as described above can be practiced using steps in a different order and / or with hardware elements in different configurations than those disclosed. Thus, while the invention has been described based on these preferred embodiments, it will be apparent to those skilled in the art that certain modifications, variations, and alternative configurations will become apparent while remaining within the spirit and scope of the invention. Accordingly, reference should be made to the appended claims to determine the scope of the invention.
Claims
1. 1. A computer program for performing artificial intelligence (AI) / machine learning (ML) model drift detection and correction for robotic process automation (RPA), stored on a non-transitory computer readable medium, the computer program comprising: at least one processor for: Analyzing information relating to input data for an AI / ML model to determine whether data drift has occurred, analyzing information relating to results from performance of the AI / ML model to determine whether model drift has occurred, or both; based on the analysis of the information, a change condition is found and if a change threshold is met and / or exceeded, then retraining the AI / ML model; The information includes one or more statistical distributions of predictions made by the AI / ML model, and the computer program further comprises: A computer program configured to map the one or more statistical distributions to respective actions taken by a plurality of RPA robots.
2. The program further comprises the at least one processor: The computer program product of claim 1 , configured to verify whether the change condition, the change threshold, or both, are met or exceeded for at least a predetermined period of time.
3. The computer program of claim 1 , wherein the analysis comprises measuring one or more distances between one-dimensional distributions of the original and new datasets for covariate drift.
4. The computer program product of claim 1 , wherein the analyzing comprises determining whether a statistical moment changes.
5. The computer program product of claim 4 , wherein the statistical moments include a mean, a variance, a skewness, a kurtosis, a covariance, or a combination thereof.
6. 2. The computer program product of claim 1 , wherein the information includes what predictions the AI / ML model made, how many predictions the AI / ML model made and how many predictions were used by an RPA robot, one or more statistical distributions of the predictions made by the AI / ML model, input data provided to the AI / ML model, or a combination thereof.
7. The computer program product of claim 1 , wherein the action is performed at each activity of an RPA workflow of each of the RPA robots.
8. The computer program further comprises: Verifying that the retrained AI / ML model meets one or more performance thresholds; 2. The computer program product of claim 1 , configured to deploy the retrained AI / ML model if the retrained AI / ML model meets the one or more performance thresholds.
9. The retrained AI / ML model is deployed together with a previous version of the AI / ML model, the program further comprising: Using real data, measure model drift of the retrained AI / ML model and a previous version of the AI / ML model; 9. The computer program product of claim 8, configured to use the retrained AI / ML model instead of a previous version of the AI / ML model if the retrained AI / ML model performs better on the real data than a previous version of the AI / ML model based on the measured model drift.
10. If the retrained AI / ML model does not perform better on the real data than a previous version of the AI / ML model based on the measured model drift, the computer program further comprises: The computer program product of claim 9 , configured to discard the retrained AI / ML model or to perform further retraining.
11. 9. The computer program product of claim 8, wherein the retrained AI / ML model is deployed using a canary testing technique or a champion / challenger testing technique.
12. The computer program further comprises: receiving, from an RPA robot, a request to execute the AI / ML model and input information regarding the AI / ML model; Verifying that the input information fits into a framework for the AI / ML model; If the input information fits into the framework, performing the AI / ML model using the input information; and The computer program product of claim 1 , configured to return a result of the execution of the AI / ML model to the RPA robot.
13. If the input information does not fit the framework, the computer program further comprises:
13. The computer program product of claim 12, configured to provide an error to the RPA robot indicating that the input information does not conform to the framework.
14. The computer program product of claim 12 , wherein the framework is constrained for multiple AI / ML models.
15. The computer program further comprises: performing a dimensionality reduction technique on the information to generate a two-dimensional representation of the information; The computer program product of claim 1 configured to display a two-dimensional representation of the information.
16. 2. The computer program product of claim 1, wherein the analysis of the information is performed across a plurality of RPA processes performed by a plurality of respective RPA robots.
17. The computer program further comprises: selecting the retrained AI / ML model or a previous version of the AI / ML model based on the input data; The computer program product of claim 1 configured to invoke a selected version of the AI / ML model.
18. The computer program product of claim 17 , wherein the selection is performed hierarchically.
19. 1. A computer-implemented method for performing artificial intelligence (AI) / machine learning (ML) model drift detection and correction for robotic process automation (RPA), comprising: Analyzing information relating to input data for an AI / ML model to determine whether data drift has occurred, analyzing information relating to results from performance of the AI / ML model to determine whether model drift has occurred, or both; If, based on the analysis of the information, a change condition is found and a change threshold is met or exceeded, or both, retraining the AI / ML model; Verifying that the retrained AI / ML model meets one or more performance thresholds; deploying the retrained AI / ML model if the retrained AI / ML model meets the one or more performance thresholds; The information includes one or more statistical distributions of predictions made by the AI / ML model, and the method further comprises: A computer-implemented method comprising mapping the one or more statistical distributions to respective actions taken by a plurality of RPA robots.
20. 20. The computer-implemented method of claim 19, further comprising verifying whether the change condition, the change threshold, or both, are met or exceeded for at least a predetermined period of time.
21. 20. The computer-implemented method of claim 19, wherein the action is taken at a respective activity of an RPA workflow of a respective RPA robot.
22. The retrained AI / ML model is deployed together with a previous version of the AI / ML model, and the method further comprises: Using real data, measure model drift of the retrained AI / ML model and a previous version of the AI / ML model; 20. The computer-implemented method of claim 19, comprising using the retrained AI / ML model instead of a previous version of the AI / ML model if the retrained AI / ML model performs better on the real data than a previous version of the AI / ML model based on the measured model drift.
23. receiving, from an RPA robot, a request to execute the AI / ML model and input information regarding the AI / ML model; Verifying that the input information fits into a framework for the AI / ML model; If the input information fits into the framework, performing the AI / ML model using the input information; and and returning a result of the execution of the AI / ML model to the RPA robot.
20. The computer-implemented method of claim 19, wherein the framework is constrained for multiple AI / ML models.
24. If the input information does not conform to the framework, the method further comprises:
24. The computer-implemented method of claim 23, comprising providing an error to the RPA robot indicating that the input information does not conform to the framework.
25. performing a dimensionality reduction technique on the information to generate a two-dimensional representation of the information; 20. The computer-implemented method of claim 19, further comprising displaying a two-dimensional representation of the information.
26. 20. The computer-implemented method of claim 19, wherein the analysis of the information is performed across a plurality of RPA processes performed by a plurality of respective RPA robots.
27. selecting the retrained AI / ML model or a previous version of the AI / ML model based on the input data; 20. The computer-implemented method of claim 19, further comprising invoking a selected version of the AI / ML model.
28. 1. A computer-implemented method for performing artificial intelligence (AI) / machine learning (ML) model drift detection and correction for robotic process automation (RPA), comprising: analyzing information including one or more statistical distributions of predictions made by the AI / ML model, the information relating to input data for the AI / ML model to determine whether data drift has occurred, results from performance of the AI / ML model to determine whether model drift has occurred, or both; Mapping the one or more statistical distributions to respective actions taken by a plurality of RPA robots; and retraining the AI / ML model if, based on the analysis of the information, a change condition is found for at least a predetermined period of time and a change threshold is met or exceeded, or both. The computer-implemented method, wherein the actions are performed at respective activities of an RPA workflow of each of the RPA robots.
29. 30. The computer-implemented method of claim 28, wherein the information includes how many predictions the AI / ML model made and how many predictions were used by the plurality of RPA robots.
30. receiving, from one RPA robot of the plurality of RPA robots, a request to execute the AI / ML model and input information regarding the AI / ML model; Verifying that the input information fits into a framework for the AI / ML model; If the input information fits into the framework, performing the AI / ML model using the input information; and and returning a result of the execution of the AI / ML model to the RPA robot.
30. The computer-implemented method of claim 28, wherein the framework is constrained for multiple AI / ML models.
Citation Information
Patent Citations
Predictive drift detection and correction
US20170330109A1
Artificial intelligence and robotic process automation for automated data management
US20190180746A1
