Provable Robustness Against a Union of _0 Adversarial Attacks

Code

• github.com/zaydh/feature-partition
  OfficialIn paperpytorch★ 0
• github.com/zaydh/target_identification
  pytorch★ 9

Abstract

Sparse or _0 adversarial attacks arbitrarily perturb an unknown subset of the features. _0 robustness analysis is particularly well-suited for heterogeneous (tabular) data where features have different types or scales. State-of-the-art _0 certified defenses are based on randomized smoothing and apply to evasion attacks only. This paper proposes feature partition aggregation (FPA) -- a certified defense against the union of _0 evasion, backdoor, and poisoning attacks. FPA generates its stronger robustness guarantees via an ensemble whose submodels are trained on disjoint feature sets. Compared to state-of-the-art _0 defenses, FPA is up to 3,000 faster and provides larger median robustness guarantees (e.g., median certificates of 13 pixels over 10 for CIFAR10, 12 pixels over 10 for MNIST, 4 features over 1 for Weather, and 3 features over 1 for Ames), meaning FPA provides the additional dimensions of robustness essentially for free.

Reproductions