%0 Conference Proceedings %T Randomized Smoothing under Attack: How Good is it in Pratice? %+ Creating and exploiting explicit links between multimedia fragments (LinkMedia) %+ the World Is Distributed Exploring the tension between scale and coordination (WIDE) %A Maho, Thibault %A Furon, Teddy %A Le Merrer, Erwan %< avec comité de lecture %B ICASSP 2022 - IEEE International Conference on Acoustics, Speech and Signal Processing %C Singapore, Singapore %I IEEE %P 1-5 %8 2022-05-22 %D 2022 %K Classifiers %K black-box attacks %K randomized smoothing %K randomized adversarial examples %Z Computer Science [cs]/Signal and Image ProcessingConference papers %X Randomized smoothing is a recent and celebrated solution to certify the robustness of any classifier. While it indeed provides a theoretical robustness against adversarial attacks, the dimensionality of current classifiers necessarily imposes Monte Carlo approaches for its application in practice. This paper questions the effectiveness of randomized smoothing as a defense, against state of the art black-box attacks. This is a novel perspective, as previous research works considered the certification as an unquestionable guarantee. We first formally highlight the mismatch between a theoretical certification and the practice of attacks on classifiers. We then perform attacks on randomized smoothing as a defense. Our main observation is that there is a major mismatch in the settings of the RS for obtaining high certified robustness or when defeating black box attacks while preserving the classifier accuracy. %G English %2 https://inria.hal.science/hal-03591421/document %2 https://inria.hal.science/hal-03591421/file/main.pdf %L hal-03591421 %U https://inria.hal.science/hal-03591421 %~ UNIV-RENNES1 %~ CNRS %~ INRIA %~ UNIV-UBS %~ INSA-RENNES %~ INRIA-RENNES %~ IRISA %~ IRISA_SET %~ INRIA_TEST %~ TESTALAIN1 %~ CENTRALESUPELEC %~ INRIA2 %~ UR1-HAL %~ UR1-MATH-STIC %~ UR1-UFR-ISTIC %~ TEST-UR-CSS %~ UNIV-RENNES %~ INRIA-RENGRE %~ INRIA-300009 %~ ANR %~ UR1-MATH-NUM