The Odds are Odd: A Statistical Test for Detecting Adversarial Examples
https://arxiv.org/abs/1902.04818
Abstract:
We investigate conditions under which test statistics exist that can reliably detect examples, which have been adversarially manipulated in a white-box attack. These statistics can be easily computed and calibrated by randomly corrupting inputs. They exploit certain anomalies that adversarial attacks introduce, in particular if they follow the paradigm of choosing perturbations optimally under p-norm constraints. Access to the log-odds is the only requirement to defend models. We justify our approach empirically, but also provide conditions under which detectability via the suggested test statistics is guaranteed to be effective. In our experiments, we show that it is even possible to correct test time predictions for adversarial attacks with high accuracy.
Authors:
Kevin Roth, Yannic Kilcher, Thomas Hofmann
Видео The Odds are Odd: A Statistical Test for Detecting Adversarial Examples канала Yannic Kilcher
Abstract:
We investigate conditions under which test statistics exist that can reliably detect examples, which have been adversarially manipulated in a white-box attack. These statistics can be easily computed and calibrated by randomly corrupting inputs. They exploit certain anomalies that adversarial attacks introduce, in particular if they follow the paradigm of choosing perturbations optimally under p-norm constraints. Access to the log-odds is the only requirement to defend models. We justify our approach empirically, but also provide conditions under which detectability via the suggested test statistics is guaranteed to be effective. In our experiments, we show that it is even possible to correct test time predictions for adversarial attacks with high accuracy.
Authors:
Kevin Roth, Yannic Kilcher, Thomas Hofmann
Видео The Odds are Odd: A Statistical Test for Detecting Adversarial Examples канала Yannic Kilcher
Показать
Комментарии отсутствуют
Информация о видео
Другие видео канала
How random is a coin toss? - NumberphileOdds and Log(Odds), Clearly Explained!!!Statistical Tests: Choosing which statistical test to useAudio Adversarial Examples: Targeted Attacks on Speech-to-TextOdds Ratios and Log(Odds Ratios), Clearly Explained!!![ML News] Google introduces Pathways | OpenAI solves Math Problems | Meta goes First Person[ML News] Plagiarism Case w/ Plot Twist | CLIP for video surveillance | OpenAI summarizes booksWhat If You Detonated a Nuclear Bomb In The Marianas Trench? (Science not Fantasy)P-value in statistics: Understanding the p-value and what it tells us - Statistics HelpHow statistics can be misleading - Mark LiddellTopographic VAEs learn Equivariant Capsules (Machine Learning Research Paper Explained)The science of cells that never get old | Elizabeth BlackburnUnderstand Calculus in 10 MinutesWhy graphene hasn’t taken over the world...yetGrokking: Generalization beyond Overfitting on small algorithmic datasets (Paper Explained)Can You Name a Country?AI learns to Create ̵K̵Z̵F̵ ̵V̵i̵d̵e̵o̵s̵ Cat Pictures: Papers in Two Minutes #1[ML News] Roomba Avoids Poop | Textless NLP | TikTok Algorithm Secrets | New Schmidhuber BlogGenerative Adversarial Networks - FUTURISTIC & FUN AI !