AI Model Weight Reduction With XAI Equivalence Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional weight reduction processing of AI models does not evaluate changes in model performance and fairness before and after the reduction, potentially leading to deteriorated equivalence.
Innovation Solution
An information processing device evaluates the equivalence between a pre-compression and post-compression model using eXplainable AI (XAI) technologies, determining the final model based on the evaluation results to ensure performance and fairness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If weight reduction processing is performed on the AI model to reduce operation amount and processor load, then productivity is improved, but reliability deteriorates due to potential changes in model equivalence and fairness
Solution Approach 1:
The patent implements feedback by evaluating equivalence between the pre-compression and post-compression models using XAI technology. The control unit compares the behavior and decisions of both models to determine if they maintain equivalent performance, providing feedback that guides whether the weight reduction preserves model reliability
Solution Approach 2:
The patent introduces XAI technology as an intermediary to assess model equivalence. This intermediary evaluation mechanism bridges the gap between weight reduction and reliability preservation by providing objective metrics to determine if the compressed model maintains the same performance characteristics as the original model
2Device complexity
If weight reduction processing is performed to reduce model size and computation, then device complexity is reduced, but measurement precision deteriorates in evaluating model performance changes
Solution Approach 1:
The patent replaces traditional mechanical performance evaluation methods with XAI-based equivalence evaluation. Instead of relying on conventional metrics that may not capture subtle performance changes, the system uses explainable AI techniques to precisely measure and compare model behavior, achieving higher measurement precision in evaluating performance changes
Data Source
AI summary
An information processing device includes a control unit. The control unit evaluates equivalence between a first learning model before weight reduction by a weight reduction method and a second learning model after the weight reduction by the weight reduction method. The control unit determines the second learning model on the basis of a result of the evaluation.