Training Data Generation Program for Concept Drift
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods face difficulties in detecting and removing inappropriate training cases from old data during retraining, leading to a lowered retraining effect due to concept drift in natural language processing and other machine learning models.
Innovation Solution
A training data generation program that uses machine learning models to calculate generation scores for both old and new data, identifies contradictory cases, and generates new training data by excluding cases that lower output quality, thereby improving the retraining effect.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If old training data is included in retraining data, then the training data quantity is maintained, but the retraining effect is lowered due to concept drift
Solution Approach 1:
The patent extracts and removes contradictory training cases from the retraining data. By identifying cases where the same input produces different outputs across old and new training data, and removing these contradictory cases, the system maintains adequate training data quantity while eliminating data that harms retraining effectiveness due to concept drift
Solution Approach 2:
The patent uses a learning quality estimation device to calculate quality scores and identify contradictory cases. This feedback mechanism evaluates the training data quality and automatically removes problematic cases, creating a closed-loop system that continuously improves retraining effectiveness by eliminating data that conflicts with concept drift
2Reliability
If contradictory training cases are removed, then the retraining effect is improved, but the training data quantity is reduced
Solution Approach 1:
The patent generates synthetic training data to replace the removed contradictory cases. By creating new training samples that reflect current concept meanings without the contradictions present in old data, the system maintains training data quantity while ensuring all data supports consistent retraining objectives
3Measurement precision
If learning quality estimation is performed for all training data, then the detection precision of contradictory cases is improved, but the computational complexity increases
Solution Approach 1:
The patent segments the training data processing into distinct phases: first calculating quality scores for individual cases, then identifying contradictory cases by comparing old and new training data, and finally removing only the problematic cases. This segmentation allows precise detection of contradictory cases without needing to reprocess all data at full complexity
Data Source
AI summary
A computer-readable storage medium storing a training data generation program for causing a computer to execute processing including: acquiring a first value by inputting first data included in a plurality of pieces of first training data to a first model that is generated through machine learning based on the plurality of pieces of first training data; acquiring a second value by inputting the first data and second data included in a plurality of pieces of second training data to a second model that is generated through machine learning based on the plurality of pieces of first and second training data; comparing the first value with the second value; and generating a plurality of pieces of third training data that does not include at least a part of the first data, based on the plurality of pieces of first and second training data, according to a result of the comparison.