Maximum-Entropy Adversarial Data Augmentation for Improved Generalization and Robustness
This work addresses the problem of enhancing robustness in deep neural networks against data shifts for machine learning practitioners, representing an incremental improvement over existing adversarial augmentation techniques.
The paper tackles the challenge of generating effective adversarial perturbations for data augmentation by proposing a maximum-entropy regularization term derived from the information bottleneck principle, which improves model robustness and generalization, achieving statistically significant gains over state-of-the-art methods on three benchmarks.
Adversarial data augmentation has shown promise for training robust deep neural networks against unforeseen data shifts or corruptions. However, it is difficult to define heuristics to generate effective fictitious target distributions containing "hard" adversarial perturbations that are largely different from the source distribution. In this paper, we propose a novel and effective regularization term for adversarial data augmentation. We theoretically derive it from the information bottleneck principle, which results in a maximum-entropy formulation. Intuitively, this regularization term encourages perturbing the underlying source distribution to enlarge predictive uncertainty of the current model, so that the generated "hard" adversarial perturbations can improve the model robustness during training. Experimental results on three standard benchmarks demonstrate that our method consistently outperforms the existing state of the art by a statistically significant margin.