Machine Learning Model Retraining for Consistent Explanations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models output different explanations for the same prediction after retraining, leading to mistrust by humans due to inconsistent explanations.
Innovation Solution
An information processing apparatus and method that generate explanatory data and calculate model parameters to reduce the difference between prediction and explanation losses, ensuring consistent explanations across retraining.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a machine learning model is retrained by adding training samples, then the prediction accuracy is improved, but the explanation consistency deteriorates
Solution Approach 1:
The patent applies preliminary action by generating explanations for training samples before model retraining and using these pre-generated explanations as reference constraints during the retraining process. This ensures that the model learns to maintain explanation consistency with previously generated explanations while improving prediction accuracy on new data.
Solution Approach 2:
The patent implements feedback by comparing explanations generated during retraining with pre-generated reference explanations, and using this comparison feedback to adjust the training process. The system monitors explanation consistency and feeds this information back into the loss function to guide the model toward maintaining stable explanations while improving predictions.
2Device complexity
If a machine learning model is trained independently of human expectations, then the model training complexity is reduced, but the explanation satisfaction deteriorates
Solution Approach 1:
The patent introduces an intermediary component that bridges machine learning model training and human expectation satisfaction. This intermediary uses pre-generated explanations as a mediator between the model's independent training process and human expectations, allowing the model to be trained relatively simply while still producing explanations that satisfy human criteria through the mediating influence of reference explanations.
Data Source
AI summary
An information processing apparatus 100 of the present invention includes: an explanation generating unit 121 that generates explanatory data explaining a prediction value output by a machine learning model as a response to an input of training data; and a parameter calculating unit 122 that calculates a parameter of the machine learning model so as to reduce a prediction loss representing a degree of difference between a preset ground truth value and a prediction value output by the machine learning model as a response to the input of the training data, and to reduce an explanation loss representing a degree of unsatisfaction, by the explanatory data, of a preset criterion that the explanatory data should satisfy.