Overcoming challenges in leveraging GANs for few-shot data augmentation

TL;DR

This paper investigates the use of GANs for few-shot data augmentation, addressing training challenges and proposing semi-supervised fine-tuning to improve classification performance in low-data regimes.

Contribution

It introduces a semi-supervised fine-tuning method for GANs to enhance few-shot data augmentation and provides a rigorous empirical analysis of existing approaches.

Findings

Training GANs with few samples is challenging under supervised regimes.

Evaluation protocols significantly impact reported performance.

Semi-supervised fine-tuning improves GAN-based augmentation effectiveness.

Abstract

In this paper, we explore the use of GAN-based few-shot data augmentation as a method to improve few-shot classification performance. We perform an exploration into how a GAN can be fine-tuned for such a task (one of which is in a class-incremental manner), as well as a rigorous empirical investigation into how well these models can perform to improve few-shot classification. We identify issues related to the difficulty of training such generative models under a purely supervised regime with very few examples, as well as issues regarding the evaluation protocols of existing works. We also find that in this regime, classification accuracy is highly sensitive to how the classes of the dataset are randomly split. Therefore, we propose a semi-supervised fine-tuning approach as a more pragmatic way forward to address these problems.