Regularizing Neural Networks via Adversarial Model Perturbation

By Yaowei Zheng, Richong Zhang, Yongyi Mao. 2020

116 citations recorded by the source corpus.

Effective regularization techniques are highly desired in deep learning for alleviating overfitting and improving generalization. This work proposes a new regularization scheme, based on the understanding that the flat local minima of the empirical risk cause the model to generalize better. This scheme is referred to as adversarial model perturbation (AMP), where instead of directly minimizing the empirical risk, an alternative "AMP loss" is minimized via SGD. Specifically, the AMP loss is obtained from the empirical risk by applying the "worst" norm-bounded perturbation on each point in the parameter space. Comparing with most existing regularization schemes, AMP has strong theoretical justifications, in that minimizing the AMP loss can be shown theoretically to favour flat local minima of the empirical risk. Extensive experiments on various modern deep architectures establish AMP as a new state of the art among regularization schemes. Our code is available at https://github.com/hiyouga/AMP-Regularizer.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Continual Learning by Modeling Intra-Class Variation (2022, background): “Despite focusing on different applications, WAP has intrinsic connections to (Wu et al., 2020; Zheng et al., 2021) where adversarial weight perturbation is shown to converge to flat minima (Hochreiter & Schmidhuber, 1997) and is beneficial to generalization.”
Towards Stable Test-Time Adaptation in Dynamic Wild World (2023, background): “SAM and its variants (Kwon et al., 2021; Zheng et al., 2021; Du et al., 2022; Chen et al., 2022b) have shown outstanding performance on several deep learning benchmarks.”
Sharpness-aware Quantization for Deep Neural Networks (2021, background): “Concurrent works have also been proposed to introduce adversarial weight perturbations to improve the robustness against adversarial examples [69] or generalization performance [74].”
Relating Adversarially Robust Generalization to Flat Minima (2021, methodology): “These measures are shown to be effective in predicting generalization in a recent largescale empirical study [27] and explicitly encouraging flatness during training has been shown to be successful in practice [74, 9, 35, 8, 26].”