Fair Few-Shot Class-Incremental Learning Storage Device
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In few-shot class-incremental learning (CIL) environments, conventional models suffer from severe unfairness due to differential catastrophic forgetting across small groups, leading to accuracy disparities and overall accuracy equality issues.
Innovation Solution
A fair few-shot CIL model is proposed, which constructs a separate storage device using samples from high and low accuracy super classes, adjusts the number of samples based on fairness criteria, and employs a regulatory scheme to reduce catastrophic forgetting and enhance model fairness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional incremental learning is performed, then new class accuracy is improved, but catastrophic forgetting occurs and existing class accuracy deteriorates
Solution Approach 1:
The patent applies preliminary action by constructing a separate storage device before incremental learning begins, pre-populating it with samples from existing classes. This advance preparation ensures that when new classes are learned, the existing class information is already preserved in the storage device, preventing catastrophic forgetting while allowing new class accuracy to improve
Solution Approach 2:
The patent introduces a separate storage device as an intermediary between the neural network model and the learning process. This storage device acts as a mediator that holds existing class samples, allowing the model to access and retain previous knowledge while learning new classes, thus resolving the conflict between plasticity (learning new information) and stability (retaining old information)
2Manufacturing precision
If fairness is prioritized in few-shot CIL, then accuracy equality across small groups is improved, but overall model accuracy may be reduced
Solution Approach 1:
The patent applies local quality by differentiating the treatment of different super classes based on their specific accuracy characteristics. The separate storage device stores samples selectively, and the incremental learning process adjusts differently for each super class, allowing local optimization for fairness in each small group while maintaining overall model performance through targeted interventions
Data Source
AI summary
Disclosed are a method and device for fair few-shot CIL. The method includes constructing a separate storage device by using samples of a first super class having first accuracy and samples of a second super class having second accuracy lower than the first accuracy, performing incremental learning on the separate storage device, adjusting the number of samples of the first super class and the number of samples of the second super class in the separate storage device when the results of the incremental learning satisfy a fairness criterion or when the first accuracy is lower than the second accuracy in the results of the incremental learning, and performing incremental learning in a next step on the separate storage device.