Graph Structure Analysis for Interpretable Machine Learning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models, such as decision trees and neural networks, lack transparency in their intermediate layers, making it difficult for users to evaluate and understand the discriminant criteria and feature extraction processes.
Innovation Solution
A graph structure analysis apparatus and method that selects analysis and comparison target ranges within a graph structure and extracts feature representations using data related to these ranges, enabling the evaluation of learning models by re-learning data branched at branch points and providing insights into feature amounts and logical rules.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If machine learning models (decision trees, neural networks) are used for discrimination and regression analysis, then analysis accuracy and productivity are improved, but interpretability and transparency of the learning model deteriorate
Solution Approach 1:
The patent segments the learning model into a graph structure representing the decision path, where nodes and edges correspond to discriminant criteria and transitions. This segmentation makes the internal logic visible and interpretable while maintaining the model's analytical capabilities.
Solution Approach 2:
The patent introduces a graph structure as an intermediary representation between the learning model and the user. This graph serves as a mediator that translates the complex internal logic of machine learning models into an interpretable visual format, allowing users to understand the decision-making process without sacrificing analytical accuracy.
2Loss of information
If discriminant criteria are set for branches of nodes in decision trees, then interpretability is improved, but the ability to extract further feature amounts from unanticipated ranges deteriorates
Solution Approach 1:
The patent implements a dynamic graph structure that can be updated and extended based on new data and discoveries. The graph is not static but can adapt to incorporate new feature amounts and discriminant criteria, allowing the system to maintain interpretability while continuously improving its feature extraction capabilities through re-learning processes.
Data Source
AI summary
A graph structure analysis apparatus 10 is an apparatus for analyzing a graph structure. The graph structure analysis apparatus 10 includes a range selection unit 11 that selects an analysis target range in the graph structure and a comparison target range to be compared with the analysis target range, and a feature representation extraction unit 12 that extracts a feature representation from data related to the analysis target range and the comparison target range, for each of the ranges.