Machine Learning Explanation Apparatus Reducing Rule Redundancy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for providing explanatory information on inference results of machine learning models often generate redundant rules, making the information difficult to interpret.
Innovation Solution
A machine learning explanation apparatus that generates and selects rules based on training data, calculates the probability of satisfaction for each rule, and outputs explanatory information including a primary rule and relevant exceptions, reducing redundancy by identifying covering relationships and selecting rules with similar influence on the model's inference results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If multiple rules are extracted and enumerated from training data to provide comprehensive explanatory information, then the completeness of explanation is improved, but the redundancy and difficulty of interpretation increase
Solution Approach 1:
The patent extracts only the most relevant and non-redundant rules from the set of all possible rules. By selectively taking out essential rules that provide the most explanatory value while removing redundant ones, the system maintains completeness of explanation while improving interpretability. This is achieved through algorithms that identify and retain only the necessary rules for accurate model explanation.
Solution Approach 2:
The patent applies different selection criteria and weighting to different rules based on their local importance and relevance to specific inference cases. Rather than treating all rules uniformly, the system adjusts the quality and detail of explanation locally according to which rules are most pertinent to the specific prediction being explained, thereby reducing redundancy while maintaining necessary completeness.
2Loss of information
If all extracted rules are presented to users for comprehensive coverage, then the completeness of information is improved, but the complexity of the explanation system increases
Solution Approach 1:
The system extracts and presents only the essential subset of rules needed for accurate explanation, removing redundant rules that would increase complexity. This selective extraction maintains information completeness while simplifying the explanation system by eliminating unnecessary components.
Solution Approach 2:
Instead of starting with all possible rules and trying to include everything, the system inverts the approach by starting with no rules and selectively adding only those that provide necessary explanatory power. This inversion reduces system complexity while maintaining completeness by building up only what is essential.
3Loss of information
If redundant rules are included in explanatory information, then the completeness of coverage is improved, but the ease of interpretation deteriorates
Solution Approach 1:
The patent systematically identifies and extracts redundant rules from the complete rule set, removing duplicates and overlapping explanations while preserving the essential coverage. This extraction process maintains completeness of information by retaining all unique explanatory content while eliminating redundant elements that hinder interpretation.
Solution Approach 2:
The system discards redundant rules that provide no additional explanatory value, while recovering and preserving the essential information content they might contain. By analyzing the semantic content of each rule, the system keeps the necessary information while discarding the redundant packaging, thereby improving interpretability without losing coverage.
Data Source
AI summary
A machine learning explanation apparatus generates rules each including a condition and a conclusion for a case where the condition is satisfied, based on training data used for training of a machine learning model, extracts, in the generated rules, a set of rules {Ri} of a family of subsets that is to be a cover of a certain rule K based on the training data, selects rules Ri for which the difference between confidence (conf) of rule K and confidence (conf) of rule Ri is less than a predetermined threshold, and outputs, for an inference result of the machine learning model, explanatory information including rule K and rule Ri remaining after deleting the selected rules Ri among the set of rules {Ri}.