Learning Data Selection Using Influence Scores Across Predictors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing supervised learning methods for generating AI models face challenges in improving prediction accuracy due to the dependency of learning data selection on specific conditions, leading to inconsistent predictor behavior and potential degradation in performance.
Innovation Solution
A method and system for selecting learning data that calculates an influence degree independent of the predictor's learning conditions, using an index to improve prediction accuracy by associating predictors with their influence scores and selecting data based on these scores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If learning data is selected based on influence degree calculated using a predictor learned under specific learning conditions, then the selection process is simplified, but the prediction accuracy may not be improved because the influence degree depends on the characteristics of the predictor
Solution Approach 1:
The patent introduces test data as an intermediary element to calculate influence degree. Instead of directly using the predictor's characteristics, the system evaluates how each learning data point influences the predictor's performance on independent test data, thereby mediating between the learning process and the selection criterion
Solution Approach 2:
The patent segments the evaluation process by calculating influence degree for each individual learning data point separately against multiple test data points. This segmentation allows the system to identify and select specific high-influence data points rather than treating the learning dataset as a homogeneous whole
2Adaptability or versatility
If a predictor learned under freely selected learning conditions is used to calculate influence degree, then the learning process is flexible, but the prediction accuracy may deteriorate because the influence degree depends on the learning conditions
Solution Approach 1:
The patent creates an equipotential evaluation framework where the influence degree calculation is made independent of the specific learning conditions. By using test data performance as the common reference point, the system equalizes the evaluation basis across different learning configurations, making the selection criterion robust to variations in learning conditions
Data Source
AI summary
A computer system accurately selects learning data for improving a prediction accuracy of a predictor, and is connected to a database that stores a plurality of pieces of learning data and information for managing a plurality of predictors generated under different learning conditions. A target predictor is selected, an influence degree representing strength of an influence of the learning data on a prediction accuracy of the target predictor for test data is calculated for each of a plurality of pieces of test data, an influence score of the learning data is calculated for the plurality of predictors based on a plurality of influence degrees of the learning data associated with the predictors, and the learning data to be used is selected from the plurality of pieces of learning data on the basis of a plurality of the influence scores of each of the plurality of pieces of learning data.