AI Model Validation via Feedback Loop Entities

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The European AI Regulation proposes to include Cellular Infrastructure as 'High Risk' systems, necessitating compliance with specific requirements, particularly in the context of multi-stakeholder AI systems where different entities control various AI-related actions such as training data provision, learning, and inferencing, without a comprehensive feedback loop to ensure adherence to the AI Act's essential requirements.

Innovation Solution

Implementing a certificate-based approach with a feedback loop between processing entities to validate compliance and report violations, using Feed-back Analysis Entities (FBAEs) and Processing Verification Entities (PVEs) to ensure that all actions meet the AI Act's requirements, including Memoryless and Memory-based FBAE approaches for corrective actions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a multi-stakeholder AI system is implemented where different entities control various AI-related actions, then the system can achieve greater adaptability and division of labor, but it becomes difficult to ensure comprehensive compliance with AI Act requirements across all stakeholders

Engineering Contradiction:
Improvemulti-stakeholder flexibilityVSAvoidcompliance assurance
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent implements a feedback loop where output metadata from one stakeholder's AI processing is verified by input verification entities of subsequent stakeholders. Each entity receives feedback about compliance status and can trigger corrective actions, ensuring that compliance requirements are maintained across the distributed multi-stakeholder system while preserving each entity's operational independence

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent introduces intermediary entities (input verification entities and output metadata entities) that act as mediators between different stakeholders. These intermediaries verify compliance without directly controlling the AI processing, enabling independent stakeholders to cooperate while maintaining compliance assurance through the intermediary verification layer

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If comprehensive verification of AI system compliance is implemented across all processing entities, then reliability and compliance assurance are improved, but the complexity of the system increases due to multiple verification and feedback mechanisms

Engineering Contradiction:
Improvecompliance verificationVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the verification function into distinct modular entities: output metadata entities that annotate processing results, and input verification entities that check compliance. Each entity has a specific, simplified responsibility rather than requiring each processing entity to implement comprehensive verification logic, reducing overall system complexity while maintaining thorough compliance checking

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent creates universal verification entities that can operate across different stakeholders and AI processing types. The input verification entities and output metadata entities serve multiple functions: verifying compliance, triggering feedback loops, and enabling corrective actions. This multi-functionality reduces the need for separate verification mechanisms for each stakeholder, managing complexity through standardized universal components

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4575838A1Artificial intelligence (AI) model validity
Publication Date: 2025.06.25 INTEL CORP
  • EP4575838A1 patent drawingFigure 1A
  • EP4575838A1 patent drawingFigure 1B
  • EP4575838A1 patent drawingFigure 2

AI summary

Systems, apparatuses, methods, and computer-readable media are provided for validation or correction of an artificial intelligence (AI) model. Specifically, embodiments relate to one or more logical entities that may be associated with a processing entity of the AI system. The logical entities may be configured to analyze, validate, and or correct one or more aspects of the AI model. Other embodiments may be described and/or claimed.