Unlocking Deterministic Robustness Certification on ImageNet

TL;DR

This paper introduces a new residual network architecture and a novel loss function to improve deterministic robustness certification for large-scale image models, achieving state-of-the-art results on multiple datasets including ImageNet.

Contribution

The paper proposes the Linear ResNet architecture and EMMA loss function, enabling scalable deterministic robustness certification on large models like ResNet and ViT for ImageNet.

Findings

Achieved state-of-the-art robust accuracy on CIFAR-10/100 and Tiny-ImageNet.

First demonstration of scalable robustness certification on ImageNet.

Developed efficient Lipschitz bound estimation for residual blocks.

Abstract

Despite the promise of Lipschitz-based methods for provably-robust deep learning with deterministic guarantees, current state-of-the-art results are limited to feed-forward Convolutional Networks (ConvNets) on low-dimensional data, such as CIFAR-10. This paper investigates strategies for expanding certifiably robust training to larger, deeper models. A key challenge in certifying deep networks is efficient calculation of the Lipschitz bound for residual blocks found in ResNet and ViT architectures. We show that fast ways of bounding the Lipschitz constant for conventional ResNets are loose, and show how to address this by designing a new residual block, leading to the \emph{Linear ResNet} (LiResNet) architecture. We then introduce \emph{Efficient Margin MAximization} (EMMA), a loss function that stabilizes robust training by simultaneously penalizing worst-case adversarial examples from \emph{all} classes. Together, these contributions yield new \emph{state-of-the-art} robust accuracy on CIFAR-10/100 and Tiny-ImageNet under $\ell_2$ perturbations. Moreover, for the first time, we are able to scale up fast deterministic robustness guarantees to ImageNet, demonstrating that this approach to robust learning can be applied to real-world applications. We release our code on Github: \url{https://github.com/klasleino/gloro}.