Machine Learning Data Augmentation for Concept Drift
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing automatic recovery techniques for machine learning models face challenges in suppressing accuracy degradation due to concept drift, particularly when the high-density region used for retraining is small, resulting in insufficient retraining data and difficulty in maintaining classification accuracy.
Innovation Solution
The proposed solution involves applying data augmentation to both dense and non-dense regions of operation data and propagating labels to augmentation data outside the cluster in a projective space, thereby increasing the scale of clusters and enhancing the number of retraining data points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data augmentation is applied only to high-density regions for retraining, then classification accuracy is maintained in stable conditions, but accuracy degradation occurs when concept drift changes data distribution
Solution Approach 1:
The patent implements dynamic selection of data regions for augmentation based on concept drift detection. The system transitions from static high-density region selection to dynamic multi-region selection (high-density, low-density, and boundary regions) when concept drift is detected, allowing the retraining strategy to adapt to changing data distributions and maintain classification accuracy.
Solution Approach 2:
The patent changes the parameter of data region selection from fixed high-density regions to variable regions including high-density, low-density, and boundary regions. This parameter change enables the system to capture diverse data characteristics under concept drift conditions, improving adaptability while maintaining accuracy through targeted augmentation in multiple regions.
2Productivity
If the high-density region is small, then the model is computationally efficient, but the number of retraining data points is insufficient leading to accuracy degradation
Solution Approach 1:
The patent segments the data space into multiple regions (high-density regions, low-density regions, and boundary regions) and selectively applies data augmentation to each segment. This segmentation allows the system to increase the number of retraining data points by utilizing multiple segments rather than relying on a single small high-density region, thereby improving productivity through efficient multi-region processing.
Solution Approach 2:
The patent merges data from multiple regions (high-density, low-density, and boundary regions) into a unified retraining dataset. By combining data points from these different segments and applying consistent augmentation strategies, the system increases the overall quantity of retraining data while maintaining computational efficiency through coordinated processing of merged datasets.
3Quantity of substance
If data augmentation is applied to expand clusters, then the number of retraining data points increases, but data distribution separation may be compromised
Solution Approach 1:
The patent applies different data augmentation strategies to different local regions of the data space. High-density regions receive one type of augmentation while low-density and boundary regions receive different augmentation approaches. This local quality differentiation ensures that data distribution separation is maintained in each region while collectively increasing the number of retraining data points across all regions.
Solution Approach 2:
The patent applies data augmentation selectively to specific regions rather than uniformly across all data. By applying augmentation partially to high-density regions and differently to low-density and boundary regions, the system increases the number of retraining data points without excessively distorting the overall data distribution separation, thus balancing quantity and precision.
Data Source
AI summary
A recording medium stores a program for causing a computer to execute a process including: classifying data into classes based on a density of the data; performing data augmentation on first data that is positioned in a region where data which is positioned in a region of a first class and which belongs to the first class exists at a higher density than a predetermined density and on second data that is positioned in a region where the data which is positioned in the region of the first class and which belongs to the first class exists at a lower density than the predetermined density; and setting, when the first data after the data augmentation and the second data after the data augmentation overlap each other, a label that corresponds to the first class to first augmentation data, the second data, or second augmentation data.