Improve Generalization and Robustness of Neural Networks via Weight Scale Shifting Invariant Regularizations

Abstract

Using weight decay to penalize the L2 norms of weights in neural networks has been a standard training practice to regularize the complexity of networks. In this paper, we show that a family of regularizers, including weight decay, is ineffective at penalizing the intrinsic norms of weights for networks with positively homogeneous activation functions, such as linear, ReLU and max-pooling functions. As a result of homogeneity, functions specified by the networks are invariant to the shifting of weight scales between layers. The ineffective regularizers are sensitive to such shifting and thus poorly regularize the model capacity, leading to overfitting. To address this shortcoming, we propose an improved regularizer that is invariant to weight scale shifting and thus effectively constrains the intrinsic norm of a neural network. The derived regularizer is an upper bound for the input gradient of the network so minimizing the improved regularizer also benefits the adversarial robustness. Residual connections are also considered and we show that our regularizer also forms an upper bound to input gradients of such a residual network. We demonstrate the efficacy of our proposed regularizer on various datasets and neural network architectures at improving generalization and adversarial robustness.

arXiv comment14 pages, 5 figures, Accepted by ICML 2021 Workshop on Adversarial Machine Learning

References in corpus (11)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Adversarially Robust Generalization Requires More Data
• A PAC-Bayesian Approach to Spectrally-Normalized Margin Bounds for Neural Networks
• Improving the Adversarial Robustness and Interpretability of Deep Neural Networks by Regularizing their Input Gradients
• Spectral Norm Regularization for Improving the Generalizability of Deep Learning
• Rethinking Softmax Cross-Entropy Loss for Adversarial Robustness
• Sparse DNNs with Improved Adversarial Robustness
• Norm-Based Capacity Control in Neural Networks
• First-order Adversarial Vulnerability of Neural Networks and Input Dimension
• Four Things Everyone Should Know to Improve Batch Normalization
• On the Implicit Bias of Dropout

Cited by in corpus (1)

• Clustering Effect of (Linearized) Adversarial Robust Models