MetaAugment: Sample-Aware Data Augmentation Policy Learning

TL;DR

MetaAugment introduces a sample-aware data augmentation method that learns personalized policies efficiently through a meta-learning framework, improving image recognition performance on multiple benchmarks.

Contribution

It proposes a novel sample-aware augmentation policy learning approach using a policy network and meta-learning, addressing the limitations of dataset-level policies and naive per-sample policies.

Findings

Achieves superior accuracy on CIFAR-10/100, Omniglot, and ImageNet.

Provides theoretical proof and convergence rate analysis of the training method.

Demonstrates efficiency and effectiveness in learning personalized augmentation policies.

Abstract

Automated data augmentation has shown superior performance in image recognition. Existing works search for dataset-level augmentation policies without considering individual sample variations, which are likely to be sub-optimal. On the other hand, learning different policies for different samples naively could greatly increase the computing cost. In this paper, we learn a sample-aware data augmentation policy efficiently by formulating it as a sample reweighting problem. Specifically, an augmentation policy network takes a transformation and the corresponding augmented image as inputs, and outputs a weight to adjust the augmented image loss computed by a task network. At training stage, the task network minimizes the weighted losses of augmented training images, while the policy network minimizes the loss of the task network on a validation set via meta-learning. We theoretically prove the convergence of the training procedure and further derive the exact convergence rate. Superior performance is achieved on widely-used benchmarks including CIFAR-10/100, Omniglot, and ImageNet.