Clustering-Based Sampling for Defect Classification Model Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for updating trained models to classify defects in manufacturing processes face challenges when new data with different distributions emerges, leading to decreased classification accuracy due to random down-sampling of existing data, which can deteriorate performance on both new and old data.
Innovation Solution
An additional training apparatus that clusters existing training data, extracts representative first data based on cluster size, and combines it with new data to create additional training data, allowing for enhanced classification accuracy of new data while maintaining accuracy on existing data through advanced training models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If random down-sampling is applied to existing training data during additional training preparation, then the training process becomes simpler and faster, but the classification accuracy for existing data deteriorates due to loss of data distribution characteristics
Solution Approach 1:
The patent changes the sampling method from random to stratified sampling based on cluster information. By dividing existing data into clusters and sampling proportionally from each cluster, the method preserves the original data distribution characteristics while still reducing the data volume for training, thus maintaining classification accuracy while improving training efficiency.
Solution Approach 2:
The patent performs clustering analysis on the existing training data before the additional training process. This preliminary action organizes the data into meaningful groups that reflect the underlying distribution patterns, enabling subsequent sampling to preserve these patterns rather than destroying them through random selection.
2Adaptability or versatility
If additional training data is created by adding new data to existing data without selective sampling, then the model improves on new data, but the classification accuracy on existing data deteriorates due to loss of tendency and features
Solution Approach 1:
The patent applies different sampling strategies to different parts of the existing data based on their cluster membership. By sampling proportionally from each cluster rather than uniformly or randomly from the entire dataset, the method preserves the local characteristics and tendencies of each data group while incorporating new data, thus maintaining reliability on existing data while improving adaptability to new data.
Solution Approach 2:
The patent segments the existing training data into multiple clusters based on their characteristics and distribution patterns. This segmentation allows the sampling process to respect the natural groupings in the data, ensuring that each cluster's contribution to the additional training data reflects its importance and characteristics in the original dataset.
Data Source
AI summary
According to one embodiment, an additional training apparatus includes processing circuitry. The processing circuitry stores, in a memory, a plurality of pieces of existing training data in which existing data is input data and a classification of defects according to the existing data is output data, and cluster data representing clusters to which the respective pieces of existing training data belong. The processing circuitry extracts a plurality of pieces of first existing training data from the pieces of existing training data in accordance with a size of each of the clusters. The processing circuitry acquires a plurality of pieces of new training data in which new data is input data and a classification of defects according to the new data is output data.