Adversarial Auto-Augment with Label Preservation: A Representation Learning Principle Guided Approach

TL;DR

This paper introduces a label-preserving, prior-free data augmentation method that leverages intermediate representations of models to generate effective augmentations, improving performance across various learning tasks without needing domain knowledge.

Contribution

It proposes a novel augmentation approach based on representation learning principles that does not require pre-defined operations or generative models, applicable to multiple machine learning settings.

Findings

Consistently improves performance in supervised, semi-supervised, and noisy-label learning.

Effective in domains lacking prior knowledge, such as medical imaging.

Enhances existing augmentation techniques without additional generative models.

Abstract

Data augmentation is a critical contributing factor to the success of deep learning but heavily relies on prior domain knowledge which is not always available. Recent works on automatic data augmentation learn a policy to form a sequence of augmentation operations, which are still pre-defined and restricted to limited options. In this paper, we show that a prior-free autonomous data augmentation's objective can be derived from a representation learning principle that aims to preserve the minimum sufficient information of the labels. Given an example, the objective aims at creating a distant "hard positive example" as the augmentation, while still preserving the original label. We then propose a practical surrogate to the objective that can be optimized efficiently and integrated seamlessly into existing methods for a broad class of machine learning tasks, e.g., supervised, semi-supervised, and noisy-label learning. Unlike previous works, our method does not require training an extra generative model but instead leverages the intermediate layer representations of the end-task model for generating data augmentations. In experiments, we show that our method consistently brings non-trivial improvements to the three aforementioned learning tasks from both efficiency and final performance, either or not combined with strong pre-defined augmentations, e.g., on medical images when domain knowledge is unavailable and the existing augmentation techniques perform poorly. Code is available at: https://github.com/kai-wen-yang/LPA3}{https://github.com/kai-wen-yang/LPA3.