INTENTION-BASED END-TO-END EXPLAINABILITY ORCHESTRATION SYSTEM

Abstract

This invention relates to a system (1) developed to increase the transparency of AI-powered decision-making processes with an end-to-end explainability orchestration flow based on an intent-driven management approach.

Description

1 TARIFF INTENTION-BASED END-TO-END EXPLAINABILITY ORCHESTRATION SYSTEM Technical Area 5 This invention provides end-to-end accountability based on an intent-driven management approach. Increasing the transparency of AI-powered decision-making processes through orchestration flow It is related to a system that provides. Previous Technique Autonomous grid management enables energy systems to be more efficient, flexible, and secure. artificial intelligence and automation technologies to ensure that it is operated in this way These systems aim to minimize human intervention by utilizing 15 especially the integration of distributed energy sources, demand-side management, and fault detection. It can make quick and effective decisions in complex processes like these. However, these decisions... The limited explainability of how it was obtained undermines the reliability and acceptance of the system. This can negatively affect its effectiveness in autonomous decision-making mechanisms. The machine learning and deep learning models used are often "black boxes". 20 Because it operates within this structure, it's unclear what kind of decisions these systems make based on what data. Understanding is becoming more difficult. This situation is particularly relevant for security in critical energy infrastructures. This raises significant issues in terms of transparency and legal accountability. The lack of transparency is not only technical, but also ethical and governance-related. Adopting autonomous grid management brings with it risks, 25 It constitutes a significant obstacle that slows things down. 2 Therefore, end-to-end accountability based on an intent-driven management approach. Increasing the transparency of AI-powered decision-making processes through orchestration flow It appears that a system providing this is needed. Turkish patent no. TR2023 / 019322, which is included in the known state of the art, 5 In the document, it describes the tools that provide connectivity to everything in 5G and beyond network systems (New The quality of network service offered by Radio Vehicle-to-Everything (NR-V2X) can be explained. The discussion revolves around prediction using artificial intelligence. This invention involves artificial intelligence. Supported operation, network for connected and autonomous mobility scenarios on 5G and beyond networks. service quality prediction with explainable artificial intelligence (eXplainable AI (XAI)) 10 It relates to an intelligent transportation system that implements key features of the V2X and C-V2X. CAM, which has a place in the market, provides network services in various use cases. operators' quality of network service offered is data-driven and It enables the identification and prediction of data using artificial intelligence. The system's predictive QoS results are also transmitted to connected devices via API. Connected tools proactively utilize MEC (Mobile Edge Computing) servers and Identifying terrestrial and non-terrestrial network access stations (base stations and satellites) and This feature provides the ability to know the relevant V2X application requirements for the service. providing uninterrupted service and positively impacting network service reliability and quality. It is doing. 20 Brief Description of the Invention The purpose of this invention is to collect data from network elements in mobile network technologies. 25 intent-based analysis of performance measurement data and data generated by customer equipment a system developed to ensure that reporting is done in an explainable way to accomplish. 3 Detailed Description of the Invention The “Intent-Based End-to-End” technology was developed to achieve the purpose of this invention. The "Explainability Orchestration System" is shown in the attached figure; Figure 1 shows a schematic view of the system that is the subject of the invention. The parts shown in the figure are individually numbered, and the corresponding numbers correspond to these numbers. It is given below. 1. System 2. Disclosure Management Server 3. Intent Management Server 4. Library Management Server 5. Model Library Database 15 6. Explainable Artificial Intelligence (XAI) Library Database 7. Model Selection Server 8. Explainable Report Production Server 9. Model Integrator 10. Data Integrator 20 11. Explainable Artificial Intelligence (XAI) Execution Server 12. Model Performance Monitor Server 13. Cross-Linguistic Model (XLM) Based Explainable Report Generation Server 14. Database An end-to-end accountability orchestration based on an intent-driven management approach. To increase the transparency of AI-powered decision-making processes through their flow. The system in question, developed for the purpose of invention (1); 4 -Decisions made in AI-powered network management are transparent, traceable, and at least one structured to ensure that it is presented in a justifiable manner Describability management server (2), -to consist of integrated AI-powered components and to provide feedback from the operator and user. or natural language expressed intentions from senior network administrator systems 5 (purpose, goal, expectation) are taken into account and these statements are structured and made suitable for processing. at least one intent management system structured to enable its conversion into a form server (3), - as a central hub for models and techniques used in artificial intelligence processes at least 10 configured to ensure storage, indexing and management a library management server (4), - various artificial intelligence models suitable for different network scenarios and use cases at least one model structured to enable systematic storage. library database (5), - to include explainable artificial intelligence techniques and make model outputs understandable. 15 structured to provide a catalog of the methods used for its introduction at least one explainable artificial intelligence library database (6), -This can be explained by analyzing classified intentions and using the most suitable artificial intelligence model. at least one configured to enable the matching of artificial intelligence (XAI) techniques Model selection server (7), 20 - why and how decisions are made by artificial intelligence models to ensure justification and the production of transparent and understandable explanations At least one explainable report production server configured (8), -included in the explainability workflow of a previously defined AI model. 25 to ensure that it is processed and works in a way that is compatible with XAI techniques at least one model integrator configured (9), -description of input specifications, historical examples and operational conditions in production the proper collection and preprocessing of the data required for the process at least one data integrator configured to ensure its retention (10), - explanation algorithms using model output and contextual data 5 to ensure that it is run and to provide information about the reasons for the outputs. At least one explainable AI execution server configured (11), -the overall reliability of the model, independent of the production processes, to enable monitoring of consistency and decision quality over time At least one configured model performance monitor server (12), -explainability outputs of devices with different configuration nomenclatures 10 In a common language, understandable, customized and contextual for users at least one cross-section structured to ensure it is presented in an enriched form Language model-based descriptive report generation server (13), -models and inputs used in explainable AI-based network management, outputs and performance indicators in a secure, consistent and scalable manner 15 It includes at least one database (14) structured to ensure its storage. The disclosure management server (2) in the system (1) which is the subject of the invention, any using the communication protocol, intent management server (3), library management server (4), model library database (5), explainable artificial intelligence library data 20 to communicate and exchange data with the database (6) and the model selection server (7) It is configured to perform. The explainability management server (2), the processing of intentions to identify explainability needs and address these needs Determining the appropriate model and explanation technique and managing the entire process end-to-end. It is structured to provide. 25 In the system in question (1), the intent management server (3) located in the system in question, any communication using the protocol, explainability management server (2) and model selection 6 to communicate with the server (7) and exchange data It is configured. The intent management server (3) not only understands the intent, at the same time ensuring that it is properly classified and completed within the system. It is structured to ensure that it is transformed into an output that can be put into action. Intent management server (3), Large Language Models (LLM) 5 using an AI-based approach, the user freely and naturally the processing of the intention statements expressed, the contextual analysis of the text with the models in question to be understood as, to be separated into semantic units and to come in a wide variety of forms. It is structured to enable the analysis of statements against a common standard. Intent management server (3), with semantic parsing technique 10 Identifying the key components in the user's intent statement and what they want to do. What is desired (action), to what is it applied (target), and in what context or under what constraints? To ensure that definitions are made in the form of (priority, time, quality objective) It is structured. The intent management server (3) is structured with the same normalization technique. Different natural language expressions that mean something need to be made singular and defined in the system, and 15 The flow is structured to enable the reader to gain a flexible understanding of natural language. Intent management server (3), not directly expressed by the detailed description technique, but the system (1) can derive from previous requests or current network status The parameters are configured to be automatically filled in. Intent management server (3), with Retrieval-Augmented Generation (RAG) architecture, 20 With LLM, pre-processed intent is taken, enriched with external information sources, and It is structured to ensure the verification process is carried out. Intent management server (3), information-based support technique, the intention in the past in the flow Similar expressions of intent that appear in domain-specific knowledge bases (service catalogs) (and in the form of network status data) and comparison with operational documentation and 25 the process involves both retrieval and generation of information It is configured to enable its execution. Intent management server (3), The intention is a predefined action, supported by contextual classification techniques. 7 to enable mapping into categories and classifying them according to categories is configured. Intent management server (3), with reliability enhancement technique based not only on the language model, but also on past examples and current data. It is structured to enable classification. The library management server (4) in the system in question (1) has any using communication protocol, explainability management server (2), model library database (5), explainable artificial intelligence library database (6), model Communicating with the election server (7) and database (14) and exchanging data It is configured to perform. The library management server (4) is configured to perform both 10 redesigning both artificial intelligence models and explainability algorithms. It is designed to provide usability, updatability, and flexibility. The model library database (5) included in the system (1) which is the subject of the invention, any library management server (4) and model selection 15 using communication protocol to communicate with the server (7) and exchange data The model library database (5) is structured for prediction, classification and Models trained for different tasks, such as anomaly detection, are on record. the type of training data, version, accuracy rate, and usage of each model are recorded. indexing and making accessible metadata in the form of context 20 It is structured to provide this. The explainable artificial intelligence library database (6) included in the system in question (1), library management server (4) and model using any communication protocol 25 to communicate with the election server (7) and exchange data is structured. Explainable artificial intelligence library database (6), SHAP (Shapley Additive Explanations), LIME (Local Interpretable Model-agnostic various explainability techniques such as Explanations and Integrated Gradients 8 including different versions and the computational complexity and implementation time of the techniques and to enable classification according to criteria such as visualization ability It is being structured. The model selection server (7) in the system (1) which is the subject of the invention, any communication 5 using the protocol intent management server (3), library management server (4), model library database (5), explainable artificial intelligence library database (6) and to communicate with the model integrator (9) and exchange data It is structured. Model selection server (7), matching process, traffic The operational scenario to which the intention belongs, in the form of optimization and failure analysis, 10 the model's accuracy and reliability history, regulation or operator request Explainability requirement level, calculation in the form of CPU / GPU usage. the current status of resources and time constraints such as the disclosure delay limit dynamic implementation based on factors such as these, and the selection of the model being more... 15 (This refers to the implementation of an adaptive selection strategy, taking into account their previous performance.) It is structured to provide this. The explainable report generation server (8) in the system (1) which is the subject of the invention, any Model integrator (9), data integrator (10), using a communication protocol, Explainable AI execution server (11), model performance monitor server 20 (12) and communication with the cross-language model-based explainable report generation server (13) It is structured to establish and carry out data exchange. (Explainable) report production server (8), operators, system administrators and regulation the ability of supervisors to oversee artificial intelligence (AI) decision-making processes and with confidence It is structured to enable its use. Explainable report production 25 server (8), explanations in different contexts such as source, service and business level enabling the creation of a multi-level workflow where the flow works synchronously across these layers. It is structured to provide explainability. 9 Model integrator (9) in the system (1) which is the subject of the invention, any communication Report generation server (8), model selection server using the protocol (7) and communicate with the explainable AI execution server (11) and data It is structured to carry out the purchase. Model integrator (9), model 5 compatibility is ensured by the input-output structure and operating mode of the selected model. and aligning data expectations, different modular working techniques Support for various types of artificial intelligence models, deep learning models in the flow, decision integrating various model types in the form of trees and probabilistic models to enable tracking and traceability using time stamps and 10 for entries and exits. By adding tracking information in the form of user tags, the decisions made by the model can be monitored. It is structured to enable this. The data integrator (10) in the system (1) which is the subject of the invention, any communication Using the protocol, the report generation server (8) can be explained, the artificial 15 Communicating and exchanging data with the intelligence execution server (11) and database (14) It is structured to perform. The data integrator (10) provides the necessary data. data should be obtained from relevant data sources, and data from different sources should be explained. normalizing and synchronizing the engines according to their expectations and For services requiring particularly low latency, the data feed is dynamically 20 It is structured to ensure its implementation. The explainable artificial intelligence execution server (11) included in the system (1) that is the subject of the invention, Report generation server (8), which can be explained using any communication protocol, Model performance monitor server (12) and cross-language model based explainable 25 to communicate with the report generation server (13) and exchange data is being configured. Explainable AI execution server (11), SHAP, LIME Among the explainability algorithms such as Integrated Gradients, the most suitable one is... the option to select automatically or via operator preference, signal for source level power change, increased latency for service level and cost impact for business level This involves generating differentiated explanations based on the level of impact of such decisions. The selected XAI technique is integrated in such a way that it can function independently of the model used. 5 It is structured to provide this. Model performance monitor server (12) included in the system (1) which is the subject of the invention, any The report production server (8) can be described using a communication protocol and Communicating and exchanging data with the explainable AI execution server (11) 10 It is configured to perform. Model performance monitor server (12), This involves analyzing whether the model makes similar decisions under similar conditions. If there are any discrepancies, they should be recorded and alarm mechanisms should be triggered. monitoring changes in data distribution and an increase in the rate of incorrect decisions. and based on the explanations provided, the model's instability or excessive sensitivity is 15 It is structured to enable the analysis of situations such as these. The invention involves the production of an explainable report based on a cross-language model in the system (1). server (13), report generation can be explained using any communication protocol. to communicate with server (8) and explainable AI execution server (11) and 20 It is structured to facilitate data exchange. Based on a cross-language model. Explainable report generation server (13), explainability in the form of SHAP and LIME The outputs of these techniques are natural and technical in different languages ​​via the XLM architecture. converting them into highly accurate descriptions, different operations and devices the production of a common text of non-common terms across providers, operator, last 25 By recognizing different user profiles such as user, administrator, or regulator The content level and language should be determined according to this profile, while the explanatory language is being translated. preserving the technical context, thanks to XLM models, allows for word-for-word translation. 11 not only should terminological consistency be observed, but also multilingual transparency and international Meeting regulatory compliance objectives for global network management scenarios in the 6G era. It is structured to ensure that it is made ready. The database (14) in the system in question (1) does not support any communication protocol 5 Using the model library database (5), explainable artificial intelligence library data to communicate with the database (6) and the data integrator (10) and to exchange data It is structured in such a way. The database (14) relates to artificial intelligence models. structured information, interpretable along with explainability outputs, Performance indicators (KPIs) that form the basis of operational decision-making processes and 10 to ensure that system configurations are recorded simultaneously is structured. Database (14), model monitoring, performance evaluation, for processes such as measuring the quality of explanation and retraining when necessary. It is structured to ensure the provision of a robust data infrastructure. Industrial Applicability The invention system (1) enables network technologies in mobile network technologies. performance measurement data collected from employees and customer equipment The data it produces is reported in an explainable, intent-based manner. 20 In the system in question (1), from the operator or high-level network administrator systems Incoming intentions expressed in natural language, explainability management in the system (1) the server (2) receives the information that the disclosure management server (2) receives requests can be made available to models using the intent management server (3) 25 structuring, converting to requests, LLMs text normalization, requests It performs tasks such as de-ambiguation and contextual enrichment. Structured requests are then processed into predefined categories. 12 distinguishing, structured intentions and explainability requirements are defined. requests are located on the library management server (4), model selection server (7), The most suitable artificial from the model library database (5) and the XAI library database (6) In making the selection of the intelligence model and the appropriate XAI technique, the chosen artificial intelligence model... The intelligence model is included in the explainability workflow via the model integrator (9). is being developed and its compatibility with XAI techniques is being ensured, explanation production process The data required for this is compiled appropriately by the data integrator (10), It is being pre-processed and made available for use by XAI models. is being brought in, XAI execution server (11) runs the description algorithms and It provides information on the reasons for the outputs, enabling explainable report production. 10 server (8) explains why and how decisions are made by artificial intelligence models. providing transparent and understandable explanations for the decision producing, XLM-based explainable report generation server (13), explainability their outputs in a common language for devices with different configuration nameplates 15 in a way that is understandable, customized and contextually enriched for users It ensures the presentation and explanation of the model independently of the production processes. Performance monitor server (12) assesses the overall reliability, consistency and performance of the model. It enables the monitoring of decision quality over time. Around these fundamental concepts, the subject of the invention is “Intent-Based End-to-End Explanability 20”. It is possible to develop a wide variety of applications related to the Orchestration System (1)” and the invention cannot be limited to the examples described here, but mainly to the claims as stated.

Claims

13 REQUESTS 1. End-to-end accountability based on an intent-driven management approach. Transparency of AI-powered decision-making processes through orchestration flow enabling its increase; 5 -Decisions made in AI-powered network management are transparent, traceable, and at least structured to ensure that it is presented in a justifiable manner including a disclosure management server (2) and -to consist of integrated AI-powered components and to be controlled by an operator or superior. 10 natural language intents from top-level network administrator systems These statements have been structured and made suitable for processing by taking into account (purpose, goal, expectation). at least one intention structured to enable its conversion into a form management server (3), - as a central hub for models and techniques used in artificial intelligence processes 15 structured to ensure storage, indexing and management at least one library management server (4), - various artificial intelligence systems suitable for different network scenarios and usage conditions structured to enable the systematic storage of models at least one model library database (5), - to include explainable artificial intelligence techniques and for model outputs to be understandable. 20 to provide a catalog of the methods used to make it available at least one explainable artificial intelligence library database structured (6), -by analyzing classified intentions and using the most suitable artificial intelligence model. to enable the matching of explainable artificial intelligence (XAI) techniques at least one configured model selection server (7), 25 - why and how decisions are made by artificial intelligence models justification and the production of transparent and understandable explanations at least one explainable report production server configured to provide (8), 14 -included in the explainability workflow of a previously defined AI model. to ensure that it is implemented and works in a way that is compatible with XAI techniques at least one model integrator configured (9), - explanation in the form of input specifications, historical examples, and operational conditions. the proper compilation of data required for the production process and preliminary 5 at least one data integrator configured to ensure processing (10), - explanation algorithms using model output and contextual data to enable its operation and provide information about the reasons for the outputs. at least one explainable AI execution server configured to do so (11), 10 -the overall reliability of the model, independent of the production processes, to enable monitoring of consistency and decision quality over time At least one configured model performance monitor server (12), -explainability outputs with different configuration nomenclatures 15 devices that are understandable, customized and contextual for users in a common language structured to ensure that it is presented in an enriched form a small cross-language model based explainable report production server (13) and -explainable AI-based network management models, Inputs, outputs and performance indicators are secure, consistent and at least one 20 configured to ensure scalable storage. a system characterized by a database (14) (1).

2. Using any communication protocol, intent management server (3), library management server (4), model library database (5), can be described Communication with the AI ​​library database (6) and model selection server (7) 25 Explanability structured for establishing and exchanging data. A system like the one in Claim 1, characterized by the management server (2) (1).

3. Identifying the explainability needs by processing intentions and these By determining the appropriate model and explanation technique for the needs, the entire process from end to end. Explanability management structured to enable end-to-end management A system like the one in Claim 1 or 2 characterized by its server (2) (1).

4. Disclosure management server using any communication protocol. (2) and communicate and exchange data with the model selection server (7) characterized by the intent management server (3) configured to perform a system like any of the above-mentioned requests (1).

5. Not only should the intention be understood, but it should also be correct within the system. classification, completion and activation in a way that can be put into action Intent management server configured to ensure that it turns into output (3) a system like any of the above-mentioned claims characterized by (1). 15 6. An artificial intelligence based on Large Language Models (LLM). using this approach, the user's intention is expressed freely and naturally. the processing of sentences, contextual analysis of the text with the models in question understanding, separation of semantic units and in a wide variety of forms 20 to ensure that incoming statements are analyzed according to a common standard the above characterized by the configured intent management server (3) a system like any of the requests (1).

7. User intent through semantic parsing technique 25 Define the basic components in the sentence and what is intended to be done. (action), to what it applies (target), and in what context or under what constraints. To ensure that definitions are made in the form of (priority, time, quality objective). 16 characterized by the intent management server (3) configured for a system like any of the above requests (1).

8. Using normalization techniques, different natural language expressions with the same meaning, singular and the system should be defined and the flow should have a flexible natural language understanding 5 with intent management server (3) configured to enable it to win a system like any of the above characterized demands (1).

9. Not directly expressed by the detail description technique, but the system (1) more Parameters that can be inferred from previous requests or the current network state: 10 Intent management configured to enable automatic filling. any of the above requests characterized by its server (3) such a system (1).

10. With Retrieval-Augmented Generation (RAG) architecture, 15 pre-processed with LLM. obtaining the intention, enriching it with external information sources, and verifying it. intent management structured to ensure the process is carried out any of the above requests characterized by its server (3) such a system (1).

11. With knowledge-based support techniques, the incoming intent appears in the flow in the past. similar statements of intent, domain-specific knowledge bases (service catalogs and (in the form of network status data) and operational documentation. comparison and retrieval of information in the process and re-creation 25 structured to ensure the collaborative execution of production (generation). from the above requests characterized by intent management server (3) a system like any other (1).

12. Predefined action of intention supported by contextual classification technique. mapping into categories and classifying them according to categories 30 17 characterized by the intent management server (3) configured to provide a system like any of the above requests (1).

13. With reliability enhancement techniques, not only the language model but also... To enable classification based on past examples and current data. 5 characterized by the intent management server (3) configured for a system like any of the above requests (1).

14. Disclosure management server using any communication protocol. (2), model library database (5), explainable artificial intelligence library data 10 to communicate with the base (6), model selection server (7) and database (14) and library management structured for data exchange in any of the above requests characterized by its server (4) such a system (1).

15. Re-evaluation for both artificial intelligence models and explainability algorithms. designed to provide usability, updatability, and flexibility. from the above requests characterized by the library management server (4) a system like any other (1).

16. Using any communication protocol, the library management server (4) and communicate with the model selection server (7) and exchange data. with the model library database (5) structured to perform a system like any of the above characterized demands (1).

17. For different tasks such as prediction, classification and anomaly detection. keeping a record of the trained models, the type of training data for each model, metadata such as version, accuracy rate, and usage context structured to ensure indexing and accessibility 18 from the above requests characterized by the model library database (5) a system like any other (1).

18. Using any communication protocol, the library management server (4) and communicate with the model selection server (7) and exchange data 5 explainable artificial intelligence library data structured to perform as in any of the above claims characterized by base (6) a system (1).

19. SHAP (Shapley Additive Explanations), LIME (Local Interpretable Model-10 various explanations such as (agnostic explanations) and integrated gradients. to include different versions of explainability techniques and the techniques computational complexity, execution time, and visualization capability It is structured to enable classification according to explainable criteria. The above 15 characterized by the artificial intelligence library database (6) a system like any of the requests (1).

20. Using any communication protocol, intent management server (3), library management server (4), model library database (5), can be described Communication with the AI ​​library database (6) and model integrator (9) 20 model selection structured for establishing and exchanging data any of the above requests characterized by its server (7) such a system (1).

21. The matching process, in the form of traffic optimization and fault analysis, is intended to be 25. the operational scenario in which it occurred, the model's history of accuracy and reliability, The level of explainability requirement in the form of regulation or operator request, The current state of computing resources in the form of CPU / GPU usage and based on factors such as time constraints, such as the explanation delay limit. dynamic implementation and the selected model's previous 30 19 implementing an adaptive selection strategy by evaluating their performance. characterized by the model selection server (7) configured to provide a system like any of the above requests (1).

22. Using any communication protocol, the model integrator (9), data 5 integrator (10), explainable AI execution server (11), model Performance monitor server (12) and cross-language model based explanation Communicate with the report generation server (13) and exchange data. with the explainable report production server (8) configured to perform A system like any of the above-mentioned demands characterized (1). 10 23. Artificial actions by operators, system administrators, and regulatory controllers intelligence (AI) being able to monitor and confidently use decision-making processes with the descriptive report production server (8) configured to provide A system like any of the above-mentioned demands characterized (1). 15 24. Disclosures in different contexts such as source, service and job level it can be created and the flow can work synchronously between these layers, very structured to provide a level of explainability 20 of the above requests characterized by the report production server (8) a system like any other (1).

25. Report generation that can be explained using any communication protocol. server (8), model selection server (7) and explainable artificial intelligence execution 25 to communicate with server (11) and exchange data The above is characterized by the structured model integrator (9). a system like any of the requests (1).

26. Model compatibility is ensured by matching the input and output structure of the selected model. Harmonizing working methods and data expectations, modular 30 supporting different types of artificial intelligence models through working techniques, flow deep learning models, decision trees, and probabilistic models the ability to integrate various model types and traceability labeling techniques with tracking information in the form of time stamps and user tags for entries and exits. 5 The above is characterized by the structured model integrator (9). a system like any of the requests (1).

27. Report generation that can be explained using any communication protocol. server (8), explainable AI execution server (11) and database 10 (14) structured to communicate and exchange data with any of the above requests characterized by data integrator (10) a system like one of them (1).

28. Obtaining the necessary data from relevant data sources, from different sources 15 normalization of incoming data according to the expectations of analysis engines and synchronization, especially for services that require low latency. to ensure that the data feed is performed dynamically The above is characterized by the structured data integrator (10). a system like any of the requests (1). 20 29. Report generation that can be explained using any communication protocol. server (8), model performance monitor server (12) and cross-language model Communicate with the report production server (13) and data based on the explainable report. explainable artificial intelligence structured to facilitate shopping 25 any of the above requests characterized by the execution server (11) a system like one of them (1).

30. Explainability algorithms in the form of SHAP, LIME, and Integrated Gradients. the ability to automatically or manually select the most suitable one among them, 30 21 Signal strength change for source level, latency increase for service level, and according to the level at which decisions are affected in the form of cost impact for the business level the generation of differentiated explanations, the use of the chosen XAI technique integration that can work independently of the model and explanations 5 to ensure that it is presented in both graphical and textual formats characterized by a structured explainable AI execution server (11) a system like any of the above-mentioned requests (1).

31. Report generation that can be explained using any communication protocol. Communication with server (8) and explainable AI execution server (11) 10 a model structured for establishing and exchanging data from the above requests characterized by the performance monitor server (12) a system like any other (1).

32. Analysis of whether the model makes similar decisions under similar conditions. 15 to ensure that any discrepancies are recorded and an alarm is issued. triggering mechanisms, data distribution changes and wrong decisions monitoring situations such as rate increases and analyzing the statements produced. analysis of situations such as model instability or excessive sensitivity Model Performance Monitor Server 20 configured to ensure this (12) like any of the above-mentioned claims characterized by system (1).

33. Report generation that can be explained using any communication protocol. Communication with server (8) and explainable AI execution server (11) 25 a cross-language model structured for establishing and exchanging data characterized by the report production server (13) based on explainable report a system like any of the above requests (1). 22 34. The outputs of the SHAP and LIME explainability techniques are XLM. through its architecture, it has high natural and technical accuracy in different languages. converting them into explanations, different operations and device providers the creation of a common text containing non-common terms, operator, end-user, By recognizing different user profiles such as administrator or regulator, content 5 the level and language should be determined according to this profile, while the explanatory language is being translated. preserving the technical context, thanks to XLM models, allows for word-for-word analysis. not only the translation but also terminological consistency should be considered, and it should be multilingual. By meeting the goals of transparency and international legal compliance in the 6G era 10 to ensure that global network management scenarios are ready Structured cross-language model-based explainable report generation server (13) like any of the above-mentioned claims characterized by system (1).

35. Using any communication protocol, model library database (5), 15 with explainable artificial intelligence library database (6) and data integrator (10) data structured for communication and data exchange as in any of the above claims characterized by its base (14) a system (1).

36. Explainability of structured information about artificial intelligence models. In a way that can be interpreted along with its outputs, it forms the basis of operational decision-making processes. the performance indicators (KPIs) and system configurations that make up data structured to ensure that it is recorded over time 25 as in any of the above claims characterized by base (14). a system (1).

37. Model monitoring, performance evaluation, measurement of disclosure quality, and a robust data infrastructure for processes such as retraining when needed 23 characterized by the database structured to enable its presentation (14) a system like any of the above-mentioned requests (1).

Images