English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

There has been emerging interest in using transductive learning foradversarial robustness (Goldwasser et al., NeurIPS 2020; Wu et al., ICML 2020;Wang et al., ArXiv 2021). Compared to traditional defenses, these defensemechanisms "dynamically learn" the model based on test-time input; andtheoretically, attacking these defenses reduces to solving a bileveloptimization problem, which poses difficulty in crafting adaptive attacks. Inthis paper, we examine these defense mechanisms from a principled threatanalysis perspective. We formulate and analyze threat models fortransductive-learning based defenses, and point out important subtleties. Wepropose the principle of attacking model space for solving bilevel attackobjectives, and present Greedy Model Space Attack (GMSA), an attack frameworkthat can serve as a new baseline for evaluating transductive-learning baseddefenses. Through systematic evaluation, we show that GMSA, even with weakinstantiations, can break previous transductive-learning based defenses, whichwere resilient to previous attacks, such as AutoAttack. On the positive side,we report a somewhat surprising empirical result of "transductive adversarialtraining": Adversarially retraining the model using fresh randomness at thetest time gives a significant increase in robustness against attacks weconsider.

Towards Evaluating the Robustness of Neural Networks Learned by Transduction