Artificial intelligence system and method for artificial intelligence system for information lifespan control

Abstract

An artificial intelligence system, comprising a controller configured to obtain training data, and train the artificial intelligence system based on the training data, wherein the artificial intelligence system is characterized in that the controller is further configured to obtain grabbing criteria, grab Objectives Specific data meeting the grabbing criteria from the training data by finding data meeting the grabbing criteria and delete the found data from the training data thereby creating a reduced training data partition, whereby the controller is further configured to train the artificial intelligence system based on the training data by training the artificial intelligence system based on the reduced training data partition and thereby create a base check point for the artificial intelligence system, which base check point is unaware of the Objectives Specific data.

Description

[0001] ARTIFICIAL INTELLIGENCE SYSTEM AND METHOD FOR ARTIFICIAL INTELLIGENCE SYSTEM FOR INFORMATION LIFESPAN CONTROL TECHNICAL FIELD

[0002] The present disclosure relates generally to the field of data management and more specifically, an artificial intelligence (Al) system and a method for the Al system for efficient data partitioning and base checkpoint creation in Al systems, such as a LLM system for information lifespan control.

[0003] BACKGROUND

[0004] Large Language Models (LLMs) relies on extensive training data to perform various tasks and handle large amounts of data to enable LLMs to adapt to dynamic and evolving environments. Consequently, managing the data lifecycle becomes critical, particularly for reducing unnecessary data retention and ensuring timely updates to the LLMs. However, conventional systems face significant challenges in managing outdated data, handling computational overhead, and selectively managing specific information within the models.

[0005] Certain attempts have been made to manage the data lifecycle in dynamically changing environments, including periodic data retraining, static data partitioning into predefined categories, and fine-tuning with unlearning techniques. However, such attempts often fail due to the complexities of continuously evolving data streams, overfitting to outdated data, retention of irrelevant information, generation of inaccurate or unreliable outcomes, and the like, which adversely affect data accuracy and adaptability. Thus, there exists a technical problem of how to manage the data lifecycle while maintaining overall system stability and adaptability in a dynamic environment.

[0006] Therefore, in light of the foregoing discussion, there exists a need to overcome the aforementioned drawbacks associated with the conventional methods and conventional systems for managing the data lifecycle.

[0007] SUMMARY

[0008] The present disclosure provides an artificial intelligence (Al) system and a method for information lifespan control, such as by providing a LLM system for efficiently partitioning the data and creating a base checkpoint. The present disclosure provides a solution to the existing problem of how to manage the data lifecycle while maintaining overall system stability and adaptability in a dynamic environment. An objective of the present disclosure is to provide a solution that overcomes at least partially the problems encountered in the prior art and provides an improved Al system (i.e., the LLM system) and improved method for the Al system.

[0009] One or more objectives of the present disclosure are achieved by the solutions provided in the enclosed independent claims. Advantageous implementations of the present disclosure are further defined in the dependent claims.

[0010] In one aspect, the present disclosure provides the artificial intelligence system comprising the controller configured to obtain training data and train the artificial intelligence system based on the training data. Furthermore, the controller is configured to obtain grabbing criteria and grab Objectives Specific data meeting the grabbing criteria from the training data by finding data meeting the grabbing criteria and deleting the found data from the training data thereby creating a reduced training data partition. Moreover, the controller is configured to train the Al system based on the training data by training the Al system based on the reduced training data partition and thereby creating a base checkpoint for the Al system, which base checkpoint is unaware of the Objectives Specific data.Advantageously, the Al system is configured to manage the training data by identifying and removing Objective-Specific data based on predefined grabbing criteria, thereby creating a reduced training data partition that optimizes the training process by reducing redundancy, improving data relevance, and reducing computational overhead. The Al system is configured to reduce the overall computational overhead and time consumption, which is required to update information by implementing a base checkpoint mechanism, thereby enabling selective information updates without complete model retraining. Furthermore, the grabbing criteria-based data partitioning enables the Al system to control specific information. Moreover, the Al system is configured to improve efficiency by reducing update operations while maintaining the overall stability of the Al system. Additionally, the Al system is configured to ensure compliance with regulatory requirements through selective information removal and updates, making the Al system suitable for privacy- sensitive and regulated environments where information lifecycle control is required.

[0011] In another aspect, the present disclosure provides the method for obtaining training data, and training the artificial intelligence system based on the training data, wherein the artificial intelligence system is characterized in that the method further comprises obtaining grabbing criteria, grabbing Objectives Specific data meeting the grabbing criteria from the training data by finding data meeting the grabbing criteria and delete the found data from the training data thereby creating a reduced training data partition. Moreover, the method further comprises training the artificial intelligence system based on the training data by training the artificial intelligence system based on the reduced training data partition and thereby creating a base checkpoint for the artificial intelligence system, which base checkpoint is unaware of the Objectives Specific.

[0012] The method achieves all the advantages and technical effects of the Al system of the present disclosure.

[0013] It is to be appreciated that all the aforementioned implementation forms can be combined.

[0014] It has to be noted that all devices, elements, circuitry, units, and means described in the present application could be implemented in the software or hardware elements or any combination thereof. All steps which are performed by the various entities described in the present application, as well as the functionalities described to be performed by the various entities are intended to mean that the respective entity is adapted to or configured to perform the respective steps and functionalities. Even if, in the following description of specific embodiments, a specific functionality or step to be performed by external entities is not reflected in the description of a specific detailed element of that entity that performs that specific step or functionality, it should be clear for a skilled person that these methods and functionalities can be implemented in respective software or hardware elements or any kind of combination thereof. It will be appreciated that features of the present disclosure are susceptible to being combined in various combinations without departing from the scope of the present disclosure as defined by the appended claims.

[0015] Additional aspects, advantages, features, and objects of the present disclosure would be made apparent from the drawings and the detailed description of the illustrative implementations construed in conjunction with the appended claims that follow. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The summary above, as well as the following detailed description of illustrative embodiments, is better understood when read in conjunction with the appended drawings. For the purpose of illustrating the present disclosure, exemplary constructions of the disclosure are shown in the drawings. However, the present disclosure is not limited to specific methods and instrumentalities disclosed herein. Moreover, those in the art will understand that the drawings are not to scale. Wherever possible, like elements have been indicated by identical numbers.

[0017] Embodiments of the present disclosure will now be described, by way of example only, with reference to computer program product comprising program instructions for performing the method to the following diagrams wherein:FIG. 1 is a block diagram of an artificial intelligence system configured to manage information lifespan, in accordance with an embodiment of the present disclosure;

[0018] FIG. 2 is a flowchart of a method for the artificial intelligence system for managing data lifespan, in accordance with an embodiment of the present disclosure;

[0019] FIG. 3 is an exemplary diagram that depicts a sequence of execution for implementing the controllable information lifespan in the Al system, in accordance with an embodiment of the present disclosure;

[0020] FIG. 4 is a diagram that illustrates an implementation scenario for updating the controllable information in the Al system, in accordance with an embodiment of the present disclosure;

[0021] FIG. 5 is an exemplary diagram that depicts a sequence of execution of a process for training and updating the Al system with controlled data, in accordance with an embodiment of the present disclosure;

[0022] FIG. 6 is a sequence diagram depicting an operational flow of implementing the controllable information lifespan in the Al system, in accordance with an embodiment of the present disclosure; and

[0023] FIG. 7 is a diagram that depicts an exemplary implementation of APIs implemented between the LLM training service and data supervisor environment, in accordance with an embodiment of the present disclosure.

[0024] In the accompanying drawings, an underlined number is employed to represent an item over which the underlined number is positioned or an item to which the underlined number is adjacent. A non-underlined number relates to an item identified by a line linking the non-underlined number to the item. When a number is non-underlined and accompanied by an associated arrow, the non-underlined number is used to identify a general item at which the arrow is pointing.

[0025] DETAILED DESCRIPTION OF EMBODIMENTS

[0026] The following detailed description illustrates embodiments of the present disclosure and ways in which they can be implemented. Although some modes of carrying out the present disclosure have been disclosed, those skilled in the art would recognize that other embodiments for carrying out or practicing the present disclosure are also possible.

[0027] FIG. 1 is a block diagram of an artificial intelligence (Al) system configured to manage information lifespan, in accordance with an embodiment of the present disclosure. With reference to FIG. 1, there is shown a block diagram 100 that includes an artificial intelligence (Al) system 102, a controller 104, a memory 106, and a network interface 108.

[0028] In accordance with an embodiment, the artificial intelligence system 102 is a Large-Language Model system. The artificial intelligence system 102 implements a Large Language Model (LLM), which is a deep learning architecture trained on extensive training data. As a result, the Al system 102 is highly adaptive and continuously fine-tunes itself in order to support multimodal capabilities by integrating text with images, audio, and structured data, ensuring broader usability.

[0029] The controller 104 is configured to obtain the training data and train the Al system 102 based on the training data. Examples of the controller 104 may include but are not limited to a central data processing device, a microprocessor, a microcontroller, a complex instruction set computing (CISC) processor, an application-specific integrated circuit (ASIC) processor, a reduced instruction set (RISC) processor, a very long instruction word (VLIW) processor, a state machine, and other processors or control circuitry.

[0030] The memory 106 is used to store the training data, which is obtained by the controller 104 in order to train the Al system 102.

[0031] Examples of implementation of the memory 106 may include but are not limited to, Electrically Erasable Programmable Read-Only Memory (EEPROM), Dynamic Random Access Memory (DRAM), Random Access Memory (RAM), Read-Only Memory (ROM), Hard Disk Drive (HDD), Flash memory, a Secure Digital (SD) card, Solid-State Drive (SSD), and / or CPU cache memory.The network interface 108 is used by the controller 104 with the memory 106 and the like. Examples of implementation of the network interface 108 may include but are not limited to a network interface, a computer port, a network socket, a network interface controller (NIC), and any other network interface device.

[0032] In operation, the controller 104 is configured to obtain training data. The controller 104 is configured to obtain the training data to train the Al system 102. In an implementation, the controller 104 is configured to obtain a large volume of training data that may include various types of data, such as text, documents, or other data suitable for training the Al system 102. Furthermore, the controller 104 is configured to train the Al system 102 based on the training data. In other words, the controller 104 is configured to obtain the training data and thereafter to train the Al system 102 based on the obtained training data. As a result, by training the Al system 102 based on the training data, the controller 104 is configured to reduce the overall computational cost and time for obtaining a dynamic response to changing environments.

[0033] Furthermore, the Al system 102 is characterized that the controller 104 is configured to obtain grabbing criteria, grab Objectives Specific data meeting the grabbing criteria from the training data by finding data meeting the grabbing criteria and deleting the found data from the training data, thereby creating a reduced training data partition. In an example, the extraction of the objectspecific data includes identifying relevant data based on structured identifiers, keywords, and semantic descriptions, followed by the removal of the identified data, resulting in a reduced training data partition in order to enhance data lifecycle control by isolating specific data while preserving neutral data, thereby optimizing training efficiency. Moreover, by minimizing the redundant or outdated information in the dataset, the controller 104 is configured to reduce the overall computational overhead, and data processing speed, and ensures that the Al system 102 operates with a streamlined, optimized dataset.

[0034] Furthermore, the controller 104 is configured to train the Al system 102 based on the training data by training the Al system 102 based on the reduced training data partition and thereby creating a base checkpoint for the Al system 102, which base checkpoint is unaware of the Objectives Specific data. In an implementation, the controller 104 is configured to train the Al system 102 on a filtered dataset that excludes objectives-specific information, ensuring that the BCP serves as a clean and unbiased foundation for future training iterations. As a result, the Al system 102 is configured to incorporate or remove specific data without requiring full retraining thereby improving the data training efficiency, enhanced scalability, and the ability to roll back to a neutral state if required. Additionally, the controller 104 is configured to ensure compliance with privacy and regulatory standards by maintaining strict control over the inclusion and exclusion of sensitive data, thereby enhancing the reliability and adaptability of the Al system 102.

[0035] In accordance with an embodiment, the controller 104 is further configured to train the Al system 102 based on the training data by creating a new Objectives-Specific Partition (OSP) and adding the grabbed Objectives-Specific Data (OSD) to the OSP. In other words, the controller 104 is configured to enhance the adaptability of the Al system 102 by creating a new Objectives-Specific Partition (OSP) and adding the extracted Objectives-Specific Data (OSD) to it, such as by isolating objectives-specific information from the main training dataset based on predefined grabbing criteria and storing the same separately in the OSP. As a result, the controller 104 is configured to enable efficient data management, allowing the Al system 102 to selectively incorporate, update, or remove specific data without disrupting the overall model. Additionally, the controller 104 is configured to provide an improved Al system 102 with reduced need for full retraining, optimized computational efficiency, and enhanced compliance with data privacy and security regulations. Therefore, by maintaining a structured and controlled repository for objectives-specific data, the controller 104 is configured to ensure that the Al system 102 remains adaptable, scalable, and capable of responding to dynamic data requirements while preserving the overall integrity and performance.

[0036] In accordance with an embodiment, the controller 104 is further configured to train the Base checkpoint (BCP) for the Al system 102 based on the Objectives-Specific Partition (OSP) and thereby create a trained model for the Al system 102 whichbecomes aware of the Objectives-Specific Data. In an implementation, the controller 104 is configured to enhance the adaptability of the Al system 102 by training the Base Check Point (BCP) using the Objectives-Specific Partition (OSP) to create a trained model that becomes aware of the Objectives-Specific Data (OSD), such as by establishing a neutral BCP by training the Al system 102 on general data, followed by a targeted training phase using the OSP, ensuring that the Al system 102 incorporates the necessary objectives-specific information. Moreover, the controller 104 is configured to enable controlled and incremental learning, allowing the Al system 102 to selectively integrate objectives-specific data while maintaining the ability to roll back to a clean BCP if needed. As a result, the controller 104 is configured to provide an improved Al system 102 with reduced computational overhead by avoiding full retraining, enhanced scalability, and adherence to privacy and regulatory compliance and also ensures an efficient data management, allowing the Al system 102 to dynamically update or remove objectives-specific information while preserving the overall reliability and performance of the Al system 102. In accordance with an embodiment, the controller 104 is further configured to obtain updated data corresponding to the Objectives Specific data and to utilize the updated data as the Objectives Specific data when adding the Objectives Specific data to the OSP. In an implementation, the controller 104 is configured to utilize the objectives-specific information while maintaining structured data management to train the Al system 102 in order to keep the Al system 102 compliant with evolving requirements, such as regulatory updates, domain-specific changes, or refined data models. As a result, the Al system 102 is configured to provide an improved model accuracy, enhanced compliance with privacy and security regulations, reduced computational overhead by avoiding unnecessary full retraining, and greater system reliability. By dynamically updating the OSP with fresh data, the controller 104 is configured to ensure that the Al system 102 remains efficient, scalable, and capable of adapting to a constantly changing data landscape.

[0037] In accordance with an embodiment, the updated data comprises modifications to the Objectives-Specific Data. In an implementation, the controller 104 is configured to incorporate the modifications to the Objectives-Specific Data (OSD, such as by identifying updates, replacing outdated information, and integrating the modified data into the Objectives-Specific Partition (OSP) to ensure that the Al system 102 reflects the most current and relevant objectives-specific information. As a result, the Al system 102 is configured to maintain compliance with evolving regulatory, security, and operational requirements while optimizing the performance of the Al system 102 with improved model accuracy, reduced computational overhead by avoiding full retraining, seamless adaptation to changing data landscapes, and enhanced compliance with data governance standards.

[0038] In accordance with an embodiment, the updated data comprises deletions to the Objectives-Specific Data. In an implementation, the controller 104 is configured to update the data that includes the deletion of outdated, expired, or non-compliant data from the Objectives-Specific Partition (OSP), such as by identifying and removing specific data elements based on predefined criteria, ensuring that the Al system 102 remains optimized, compliant, and free of obsolete information. As a result, the Al system 102 is configured to enhance the data governance and regulatory compliance while preventing unnecessary computational resource consumption with reduced storage and processing overhead, improved Al system 102 reliability by ensuring only relevant data is retained, enhanced data privacy and security compliance, and greater adaptability to dynamic regulatory requirements that allows the Al system 102 to maintain efficient data lifecycle management while optimizing performance and compliance.

[0039] In accordance with an embodiment, the updated data comprises substitution^) to the objectives-specific data. In other words, the Al system 102 is configured to replace the existing data elements with updated and more relevant information, ensuring that the Al system 102 consistently operates with the most accurate objectives-specific data in order to maintain data integrity while allowing the Al system 102 to dynamically adapt to evolving requirements. As a result, the Al system 102 is configured to reduce the overall computational overhead by selectively updating data instead of performing full retraining, enhancedcompliance with regulatory and security standards, and increased operational efficiency. By incorporating real-time substitutions, the Al system 102 remains up-to-date, scalable, and aligned with changing data landscapes.

[0040] In accordance with an embodiment, the controller 104 is configured to obtain new controllable data, add the new controllable data to the OSP thereby creating an updated OSP, and train the base checkpoint for the Al system 102 based on the updated OSP and thereby create an updated trained model for the artificial intelligence system, which trained model is aware of the Objectives Specific data. In other words, the controller 104 is configured to enhance the Al system 102 by incorporating new controllable data into the Objectives-Specific Partition (OSP), thereby creating an updated OSP that reflects the latest objectives-specific information, such as by identifying new relevant data, integrating it into the OSP, and training the Base Check Point (BCP) based on the updated OSP to generate an Al model that is aware of the newly introduced Objectives-Specific Data (OSD). As a result, the Al system 102 is configured to dynamically update without requiring full retraining thereby providing an improved adaptability to evolving data requirements, enhanced scalability, optimized computational efficiency by reducing redundant retraining efforts, and better compliance with privacy and regulatory standards.

[0041] Advantageously, the Al system 102 is configured to dynamically manage training data by identifying and removing Objective-Specific data based on predefined grabbing criteria, thereby creating a reduced training data partition that optimizes the training process by reducing redundancy, improving data relevance, and reducing computational overhead. The Al system 102 is configured to reduce the overall computational overhead and time consumption, which is required to update information by implementing a base checkpoint mechanism, thereby enabling selective information updates without complete model retraining. Furthermore, the grabbing criteria-based data partitioning enables the Al system 102 to control the specific information. Moreover, the Al system 102 is configured to improve efficiency by reducing update operations while maintaining the overall stability of the Al system 102. Additionally, the Al system 102 is configured to ensure compliance with regulatory requirements through selective information removal and updates, making the Al system 102 suitable for privacy-sensitive and regulated environments where information lifecycle control is required.

[0042] FIG. 2 is a flowchart of a method for an Artificial intelligence (Al) system for managing data lifespan, in accordance with an embodiment of the present disclosure. With reference to FIG. 2, there is shown a flowchart of a method 200 that includes steps 202 to 206 and sub-steps 204A to 204B. The Al system (i.e., the Al system 102 of FIG. 1) is configured to execute the method 200.

[0043] At step 202, the method 200 includes obtaining training data. Firstly, the data is obtained from various sources, such as databases, that are used to train the Al system 102 while ensuring the accuracy and reliability of the Al system 102 and training the Al system 102 based on the training data. Moreover, the Al system 102 is characterized in that the method 200 further comprises obtaining grabbing criteria, such as at sub-step 204A. Thereafter at sub-step 204B, the method 200 includes grabbing objectives-specific data meeting the grabbing criteria from the training data by finding data meeting the grabbing criteria and deleting the found data from the training data thereby creating a reduced training data partition whereby the method 200 further comprises training the artificial intelligence system 102 based on the training data by training the artificial intelligence system based on the reduced training data partition, such as at step 204 and thereby create a base checkpoint for the artificial intelligence system, which base checkpoint is unaware of the Objectives Specific, such as at step 206.

[0044] In accordance with an embodiment, the method 200 further comprises training the Al system 102 based on the training data by creating a new OSP and adding the Objectives Specific to the OSP training the base checkpoint for the artificial intelligence system based on the OSP and thereby creating a trained model for the Al system 102, which trained model is aware of the Objectives Specific data. The method 200 creates a new Objectives-Specific Partition (OSP) to add the Objectives-Specific Data to the data training partition and trains the Base Check Point (BCP) with the OSP to provide a fully trained Al system 102 by maintaining a separation between BCP and OSP. In an implementation, the controller 104 is configured to manage theobjective-specific data while preserving the integrity of the BCP and reducing computational overhead by retraining the Al system 102 with objective-specific data.

[0045] Advantageously, the method 200 is used to dynamically manage training data by identifying and removing Objective-Specific data based on predefined grabbing criteria, thereby creating a reduced training data partition that optimizes the training process by reducing redundancy, improving data relevance, and reducing computational overhead. The method 200 is used to reduce the overall computational overhead and time consumption, which is required to update information by implementing a base checkpoint mechanism thereby enabling selective information updates without complete model retraining. Furthermore, the grabbing criteria-based data partitioning enables the Al system 102 to control the specific information. Moreover, the method 200 is used to improve efficiency by reducing update operations while maintaining the overall stability of the Al system. Additionally, the method 200 is used to ensure compliance with regulatory requirements through selective information removal and updates, making the Al systeml02 suitable for privacy-sensitive and regulated environments where information lifecycle control is required. The steps 202 to 208 are only illustrative, and other alternatives can also be provided where one or more steps are added, one or more steps are removed, or one or more steps are provided in a different sequence without departing from the scope of the claim herein.

[0046] There is further provided a computer program product comprising program instructions for performing the method 200 when executed by one or more processors in the Al system 102. The computer program product is implemented as an algorithm embedded in a software stored in a non-transitory computer-readable storage medium. The non-transitory computer-readable storage means may include, but are not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. Examples of implementation of computer-readable storage medium, but are not limited to, Electrically Erasable Programmable Read-Only Memory (EEPROM), Random Access Memory (RAM), Read Only Memory (ROM), Elard Disk Drive (EIDD), Flash memory, a Secure Digital (SD) card, Solid-State Drive (SSD), a computer-readable storage medium, and / or CPU cache memory.

[0047] FIG. 3 is an exemplary diagram that depicts a sequence of execution for implementing controllable information lifespan in the Al system, in accordance with an embodiment of the present disclosure. With reference to FIG. 3, there is shown a diagram 300 that depicts the sequence of the execution for implementing the controllable information lifespan in the Al system 102.

[0048] In an implementation scenario, an initial training data source 302 provides comprehensive data that includes both general and objective-specific data to train the Al system 102. The storage component, base data 304, holds the clean and neutral data during the pre-training process while ensuring the data is used to create the Base Check Point (BCP) without including Objectives-Specific Data. Moreover, during the initial data preparation phase 318, the processing components, including BCP 306 and base checkpoint 308, perform data preparation by separating the neutral data from the Objectives-Specific Data. The system identifies and isolates Objectives-Specific Data based on the Grabbing Filter criteria provided by the Data Supervisor, which is transferred through add-on partition 310 to CTLD 312. Furthermore, the controllable data 314 is then used to train the Al system 102, incorporating the Objectives-Specific Data while maintaining BCP neutrality. In the controllable training phase 320, the trained model 316 implements rollback management capabilities, ensuring an efficient and reliable Al system 102. As a result, the Al system 102 is configured to reduce the overall computational overhead and enables selective information control through an efficient two-phase training approach with checkpoint-based rollback capability, eliminating the need for complete model retraining while maintaining model stability.

[0049] FIG. 4 is a diagram that illustrates an implementation scenario for updating the controllable information in the Al system, in accordance with an embodiment of the present disclosure. With reference to FIG. 4, there is shown a diagram 400 that depicts an implementation scenario for updating the controllable information in the Al system 102 (of FIG. 1).In an implementation scenario, the data supervisor (DS) 402 initiates a request to update or remove information from the LLM system 412. Firstly, the DS 402 provides input through the grabbing filter (GF) 404 which is processed and transferred as data (D) 406 to the LLM 412. Thereafter, when specific information needs to be updated, the DS 402 provides new data through a structured flow 408 to retrain the system from the Base Check Point (BCP). After that, the DS 402 can either provide updated Controllable Training Lifecycle Data (CTLD) 410 for retraining or specify a CTLD_ZERO flag, where CTLD_ZERO indicates that the information should be removed rather than updated. The Al system 102 then performs a rollback to the BCP, ensuring reliable removal of previous object-specific information from the model when CTLD_ZERO is specified or retrains with the new CTLD when updates are provided. As a result, the Al system 102 is configured to provide selective information updating and removal through a controlled rollback mechanism, providing reliable management of model information while maintaining system integrity and reducing computational overhead compared to traditional retraining of the Al system 102.

[0050] FIG. 5 is an exemplary diagram that depicts a sequence of execution of a process for training and updating the Al system with controlled data, in accordance with an embodiment of the present disclosure. With reference to FIG. 5, there is shown a diagram 500 comprising an initialization phase 528 and a re-occurring update phase 550.

[0051] In an exemplary scenario, the initialization phase begins when the DS 504 initiates registration for controllable data services by providing a grabbing filter (i.e., at operation 506) to LLM Service 508, which then processes the grabbed data and creates partitions (i.e., at operation 510 and operation 512). Thereafter, the cleaned and partitioned Base Data Set (BDS) is stored in LLM Config 514, after which the grabbed data is sent back to the DS (i.e., at operation 516) to retrieve fresh data from the Local Storage 502 (i.e., at operation 518), enabling the LLM Service 508 to conduct clean training on the dataset (i.e., at operation 520). Furthermore, the DS retrieves updated Controllable Training Lifecycle Data (CTLD) (i.e., at operation 522), with the Base Check Point (BCP) being saved to the LLM Config 514 (i.e., at operation 524) and the CTLD stored in the Local Storage 502 (i.e., at operation 526). During the re-occurring update phase 550, fresh data is retrieved from the Local Storage 502 (i.e., at operation 530) along with the previously stored CTLD (i.e., at operation 532), allowing the DS to prepare updated controllable data (i.e., at operation 534) and store the updated CTLD (i.e., at operation 536). The updated CTLD and BCP ID are provided to the LLM Service (i.e., at operation 538), followed by retrieval of BDS and BCP data from the LLM Config 514 (i.e., at operation 540). The LLM Service then performs arollbackto the Base checkpoint (i.e., at operation 542) before training the model with the fresh CTLD and BDS (i.e., at operation 544). Finally, the updated model is saved to the LLM Config 514 (i.e., at operation 546), and a confirmation message is sent to the DS (i.e., at operation 548). As a result, the Al system 102 is used to provide an efficient data lifecycle management through a structured two-phase approach with checkpoint-based rollback capability, enabling selective updates while maintaining model integrity and reducing computational costs.

[0052] FIG. 6 is a sequence diagram depicting an operational flow of implementing the controllable information lifespan in the Al system, in accordance with an embodiment of the present disclosure. With reference to FIG. 6, there is shown a diagram 600 depicting an operational flow of implementing the controllable information lifespan in the Al system.

[0053] In an implementation, at operation 602, the method includes data preparation to ensure data refinement. At operation 604, the method implements targeted grabbing, where specific data is identified and extracted based on grabbing criteria provided by the Data Supervisor. The system then creates data partitions for controlled training at operation 606. At operation 608, the Al system 102 is configured to pre-train the using a neutral dataset. The controllable 104 is configured to train, such as at operation 610 by creating the Base Check Point (BCP) thereby ensuring the Al system 102 remains independent of controllable data. At operation 612, add-on training is performed to incorporate controllable data while preserving the integrity of trained neutral data. The system proceeds through operations 614-620, implementing specific data controls and training refinements. At operations 622 and 624, the method includes validation and model release phases respectively, where the system ensures model accuracy and makes the validated Al system available for operational use. As a result, the Al system 102 is configured toimplement a structured controllable training approach with targeted data grabbing and partitioning, enabling efficient model updates through BCP-based rollback capability while reducing computational overhead by a factor of 100-1000 compared to traditional retraining methods.

[0054] FIG. 7 is a diagram that depicts an exemplary implementation of Application Programming Interfaces (APIs) implemented between the LLM Training Service and data supervisor environment, in accordance with an embodiment of the present disclosure. With reference to FIG. 7, there is shown a diagram 700 depicting the exemplary implementation of APIs between the LLM Training Service and Data Supervisor Environment.

[0055] In an implementation, the LLM Training Service 702 implements two primary APIs for communication with the Data Supervisor Environment 704. The first API, RegisterForControl (i.e., at operation 706 and at operation 708), enables the Data Supervisor to initiate model cleaning by providing a Grabbing Filter, which specifies criteria for accurate specific data detection and retrieval from the original training set. Upon processing, the API returns areferenceto the grabbed dataset (i.e., at operation 708) to the Data Supervisor Environment 704, effectively transferring ownership of the specific data. Furthermore, the UpdateModel API (i.e., at operation 710 and operation 712) allows the Data Supervisor Environment 704 to initiate model continual training from a specific Base Check Point (BCP). The Data Supervisor provides both the BCP_ID and the most recent Controllable Training Lifecycle Data (CTLD) as part of the update request (i.e., at operation 710), enabling controlled model updates while maintaining system integrity. The Al system 102 then processes this request and returns appropriate responses (i.e., at operation 712). As a result, the Al system 102 is configured to provide a secure and efficient API architecture that enables controlled data management and model updates through well-defined interfaces, reducing complexity while ensuring proper data ownership and update control mechanisms.

[0056] Modifications to embodiments of the present disclosure described in the foregoing are possible without departing from the scope of the present disclosure as defined by the accompanying claims. Expressions such as "including", "comprising", "incorporating", "have", "is" used to describe, and claim the present disclosure are intended to be construed in a non-exclusive manner, namely allowing for items, components or elements not explicitly described also to be present. Reference to the singular is also to be construed to relate to the plural. The word "exemplary" is used herein to mean "serving as an example, instance or illustration". Any embodiment described as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments or to exclude the incorporation of features from other embodiments. The word "optionally" is used herein to mean "is provided in some embodiments and not provided in other embodiments". It is appreciated that certain features of the present disclosure, which are, for clarity, described in the context of separate embodiments, may also be provided in combination in a single embodiment. Conversely, various features of the invention, which are, for brevity, described in the context of a single embodiment, may also be provided separately or in any suitable combination or as suitable in any other described embodiment of the disclosure.

Claims

CLAIMS1. An artificial intelligence system (102), comprising a controller (104) configured toobtain training data, andtrain the artificial intelligence system (102) based on the training data, wherein the artificial intelligence system (102) is characterized in that the controller (104) is further configured toobtain grabbing criteria,grab Objectives Specific data meeting the grabbing criteria from the training data by finding data meeting the grabbing criteria and delete the found data from the training data thereby creating a reduced training data partition, whereby the controller (104) is further configured to train the artificial intelligence system (102) based on the training data bytraining the artificial intelligence system (102) based on the reduced training data partition and thereby creating a base checkpoint for the artificial intelligence system (102), which base checkpoint is unaware of the Objectives Specific data.

2. The artificial intelligence system (102) according to claim 1, wherein the controller (104) is further configured to train the artificial intelligence system (102) based on the training data bycreating a new Objectives Specific Partition, OSP, andadding the grabbed Objectives Specific data to the OSPtraining the base checkpoint for the artificial intelligence system (102) based on the OSP and thereby create a trained model for the artificial intelligence system (102), which trained model is aware of the Objectives Specific data.

3. The artificial intelligence system (102) according to claim 2, wherein the controller (104) is further configured to obtain updated data corresponding to the Objectives Specific data and to utilize the updated data as the Objectives Specific data when adding the Objectives Specific data to the OSP.

4. The artificial intelligence system (102) according to claim 3, wherein the updated data comprises modification(s) to the Objectives Specific data.

5. The artificial intelligence system (102) according to claim 3 or 4, wherein the updated data comprises deletion(s) to the Objectives Specific data.

6. The artificial intelligence system (102) according to claim 3, 4 or 5, wherein the updated data comprises substitution!' s) to the Objectives Specific data.

7. The artificial intelligence system (102) according to any of claims 2 to 6, wherein the controller (104) is further configured toobtain new controllable data,add the new controllable data to the OSP thereby creating an updated OSP andtrain the base check point for the artificial intelligence system based on the updated OSP and thereby create an updated trained model for the artificial intelligence system (102), which trained model is aware of the Objectives Specific data.

8. The artificial intelligence system (102) according to any preceding claim, wherein the artificial intelligence system is a Large-Language Model system.

9. A method (200) for an artificial intelligence system (102)the method (200) comprising:obtaining training data,training the artificial intelligence system (102) based on the training data, wherein the artificial intelligence system (102) is characterized in that the method (200) further comprisesobtaining grabbing criteria,grabbing Objectives Specific data meeting the grabbing criteria from the training data by finding data meeting the grabbing criteria and delete the found data from the training data thereby creating a reduced training data partition, whereby the method (200) further comprises training the artificial intelligence system (102) based on the training data bytraining the artificial intelligence system (102) based on the reduced training data partition and thereby create a base check point for the artificial intelligence system (102), which base check point is unaware of the Objectives Specific.

10. The method (200) according to claim 9, wherein the method (200) further comprises training the artificial intelligence system (102) based on the training data bycreating a new OSP, andadding the Objectives Specific to the OSPtraining the base check point forthe artificial intelligence system (102)based on the OSP and thereby create a trained model for the artificial intelligence system (102), which trained model is aware of the Objectives Specific data.

11. A computer program product comprising program instructions for performing the method (200) according to claim 9 or 10, when executed by one or more processors in an artificial intelligence system (102).

Images