ML Inference Validation via SHAP Factor Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Evaluating the validity of inference results from machine learning models is challenging due to their 'black box' nature, making it difficult to verify whether the models are properly learned and if their outputs are accurate.
Innovation Solution
A method is proposed that involves extracting specific data from the inference results using predictive errors, and then performing factor analysis using explainable AI techniques like SHAP to understand the contribution of each input variable to the inference results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all input data and output inference results are verified to check model validity, then the completeness of verification is improved, but the amount of data to be verified becomes enormous making it difficult to complete
Solution Approach 1:
The patent segments the verification process by dividing all inference results into multiple groups based on predictive error magnitude. Instead of verifying all data uniformly, the system focuses factor analysis on specific segments (groups with larger predictive errors) that are most likely to reveal model issues, thereby reducing the overall verification workload while maintaining reliability.
Solution Approach 2:
The patent applies local quality by performing factor analysis selectively on specific groups of inference results rather than uniformly across all data. Groups with larger predictive errors receive detailed factor analysis to identify potential model problems, while groups with smaller errors are handled differently, optimizing verification resources where they are most needed.
2Measurement precision
If factor analysis is performed on all inference results, then the thoroughness of model evaluation is improved, but the time and computational resources required increase significantly
Solution Approach 1:
The patent segments inference results into multiple groups based on predictive error characteristics. Factor analysis is then performed selectively on specific segments (particularly those with larger errors) rather than on all inference results, significantly reducing the time and computational resources required while maintaining thorough evaluation of critical areas.
Solution Approach 2:
The patent applies partial action by performing factor analysis on only a portion of inference results - specifically those in groups with larger predictive errors - rather than conducting exhaustive analysis on all data. This partial approach is sufficient to identify model validity issues without the prohibitive cost of complete analysis.
3Adaptability or versatility
If the machine learning model is used to predict future values, then the utility of the model is improved, but the predictive error increases making validation more difficult
Solution Approach 1:
The patent uses feedback by calculating predictive errors between autoregressive model predictions and actual inference results. These error measurements feed back into the verification process, allowing the system to identify groups with larger errors and apply factor analysis selectively to those groups, thereby validating the model's utility while accounting for predictive inaccuracies.
Data Source
AI summary
There is provided an evaluation method for evaluating the validity of inference results outputted by a machine learning model. This evaluation method includes the steps of: extracting specific data from the inference results outputted by the machine learning model (Steps S11 to S13); and performing factor analysis of the machine learning model on the specific data (Steps S21 and S22). This allows the validity of the machine learning model which outputs inference results from sensor data to be efficiently checked without performing factor analysis on all inference results.