A Characterization of Semi-Supervised Adversarially-Robust PAC Learnability

Idan, Attias,; Hanneke, Steve; Mansour, Yishay

doi:10.48550/arxiv.2202.05420

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They moreover draw a more nuanced picture of proper robust learnability with access to unlabelled random examples. Attias et al (2022) also study the sample complexity of robust learning in the semi-supervised framework. Notably, in the realizable setting, their labelled sample complexity bounds are linear in a variant of the VC dimension where, for a shattered set, the perturbation region around a given point must share the same label.…”

Section: Existence Of Adversarial Examplesmentioning

confidence: 99%

Sample Complexity Bounds for Robustly Learning Decision Lists against Evasion Attacks

Hintersdorf

Struppek

Kersting

2022

Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence

View full text Add to dashboard Cite

Membership inference attacks (MIAs) aim to determine whether a specific sample was used to train a predictive model. Knowing this may indeed lead to a privacy breach. Most MIAs, however, make use of the model's prediction scores - the probability of each output given some input - following the intuition that the trained model tends to behave differently on its training data. We argue that this is a fallacy for many modern deep network architectures. Consequently, MIAs will miserably fail since overconfidence leads to high false-positive rates not only on known domains but also on out-of-distribution data and implicitly acts as a defense against MIAs. Specifically, using generative adversarial networks, we are able to produce a potentially infinite number of samples falsely classified as part of the training data. In other words, the threat of MIAs is overestimated, and less information is leaked than previously assumed. Moreover, there is actually a trade-off between the overconfidence of models and their susceptibility to MIAs: the more classifiers know when they do not know, making low confidence predictions, the more they reveal the training data.

show abstract

Section: Existence Of Adversarial Examplesmentioning

confidence: 99%