MACER: Attack-free and Scalable Robust Training via Maximizing Certified Radius

TL;DR

MACER is a fast, attack-free training method that improves certified robustness of neural networks by maximizing the certified radius, outperforming existing provable defenses in efficiency and robustness.

Contribution

MACER introduces a novel attack-free training algorithm that maximizes the certified radius, enabling scalable and efficient robust model training without adversarial attacks.

Findings

MACER achieves larger average certified radius across datasets.

MACER trains faster than adversarial methods.

MACER is effective on diverse datasets including ImageNet.

Abstract

Adversarial training is one of the most popular ways to learn robust models but is usually attack-dependent and time costly. In this paper, we propose the MACER algorithm, which learns robust models without using adversarial training but performs better than all existing provable l2-defenses. Recent work shows that randomized smoothing can be used to provide a certified l2 radius to smoothed classifiers, and our algorithm trains provably robust smoothed classifiers via MAximizing the CErtified Radius (MACER). The attack-free characteristic makes MACER faster to train and easier to optimize. In our experiments, we show that our method can be applied to modern deep neural networks on a wide range of datasets, including Cifar-10, ImageNet, MNIST, and SVHN. For all tasks, MACER spends less training time than state-of-the-art adversarial training algorithms, and the learned models achieve larger average certified radius.