Few-Shot Continual Learning via Flat-to-Wide Approaches

TL;DR

This paper introduces FLOWER, a few-shot continual learning method that finds flat-wide minima using data augmentation within a restricted sampling space, effectively reducing catastrophic forgetting in data-scarce scenarios.

Contribution

The paper proposes a novel flat-to-wide learning approach with a ball generator-based data augmentation for few-shot continual learning, addressing overfitting and catastrophic forgetting.

Findings

FLOWER outperforms prior methods on small base tasks

The approach effectively mitigates catastrophic forgetting

Data augmentation within a minimal enclosing ball improves learning

Abstract

Existing approaches on continual learning call for a lot of samples in their training processes. Such approaches are impractical for many real-world problems having limited samples because of the overfitting problem. This paper proposes a few-shot continual learning approach, termed FLat-tO-WidE AppRoach (FLOWER), where a flat-to-wide learning process finding the flat-wide minima is proposed to address the catastrophic forgetting problem. The issue of data scarcity is overcome with a data augmentation approach making use of a ball generator concept to restrict the sampling space into the smallest enclosing ball. Our numerical studies demonstrate the advantage of FLOWER achieving significantly improved performances over prior arts notably in the small base tasks. For further study, source codes of FLOWER, competitor algorithms and experimental logs are shared publicly in \url{https://github.com/anwarmaxsum/FLOWER}.