Rule Generation Apparatus Ensuring Real Example Clusters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
When a rule base model is trained using a combination of real and synthetic examples, the accuracy of the rules improves but the explanation property is reduced, making it difficult to verify the validity of the rules as clusters may consist only of synthetic examples without real examples.
Innovation Solution
A rule generation apparatus and method that acquires a training set including real and synthetic examples, and generates rule groups by dividing the examples into clusters using a rule base model, ensuring each cluster includes at least N real examples, thereby improving the explanation property of the rules.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a rule base model is trained using a combination of real and synthetic examples, then the accuracy of the rules improves, but the explanation property is reduced
Solution Approach 1:
The patent segments the training examples into two distinct types: real examples and synthetic examples. By separating these examples and applying different constraints to each type during rule generation, the system can leverage the accuracy benefits of synthetic examples while preserving the explanation property through real examples. The rule generation process is divided into stages where real examples are used to establish valid clusters and synthetic examples are used to improve accuracy within those validated clusters.
2Measurement precision
If synthetic examples are used to improve rule accuracy, then more clusters may consist only of synthetic examples without real examples, making it difficult to verify rule validity
Solution Approach 1:
The patent introduces real examples as an intermediary element that mediates between synthetic examples and rule validity verification. Real examples serve as a reference standard to verify whether clusters formed by synthetic examples are valid. The system uses real examples to establish ground truth clusters, then compares synthetic example clusters against these real clusters to verify their validity, thus enabling rule verification even when clusters contain primarily synthetic examples.
Data Source
AI summary
In a rule generation apparatus, a rule generation unit generates a rule group for dividing a training example into a plurality of clusters related to target values using a rule base model so that a “first constraint” is satisfied. The training example includes at least one real example and at least one synthetic example. Each of the real and the synthetic examples includes a feature value vector of which vector elements are one or a plurality of feature values corresponding to feature parameters different from each other, and a target value. The feature value and the target value included in each of the real examples are measured values, while each of the synthetic examples is an example formed based on the real example. The “first constraint” includes a constraint that each of the clusters includes at least N (N is a natural number) real example.