Puzzle Mix: Exploiting Saliency and Local Statistics for Optimal Mixup

TL;DR

Puzzle Mix is a novel data augmentation technique that leverages saliency and local statistics to create more effective mixed training examples, improving neural network generalization and robustness.

Contribution

It introduces a new mixup method that explicitly uses saliency and statistics, optimizing mixing masks through a novel objective, achieving state-of-the-art results.

Findings

Outperforms existing mixup methods on CIFAR-100, Tiny-ImageNet, and ImageNet.

Enhances both generalization and adversarial robustness.

Uses an optimization framework combining multi-label and transport objectives.

Abstract

While deep neural networks achieve great performance on fitting the training distribution, the learned networks are prone to overfitting and are susceptible to adversarial attacks. In this regard, a number of mixup based augmentation methods have been recently proposed. However, these approaches mainly focus on creating previously unseen virtual examples and can sometimes provide misleading supervisory signal to the network. To this end, we propose Puzzle Mix, a mixup method for explicitly utilizing the saliency information and the underlying statistics of the natural examples. This leads to an interesting optimization problem alternating between the multi-label objective for optimal mixing mask and saliency discounted optimal transport objective. Our experiments show Puzzle Mix achieves the state of the art generalization and the adversarial robustness results compared to other mixup methods on CIFAR-100, Tiny-ImageNet, and ImageNet datasets. The source code is available at https://github.com/snu-mllab/PuzzleMix.