Salvato in:
Dettagli Bibliografici
Autori principali: Prach, Bernd, Lampert, Christoph H.
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:https://arxiv.org/abs/2412.04245
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912486766149632
author Prach, Bernd
Lampert, Christoph H.
author_facet Prach, Bernd
Lampert, Christoph H.
contents Despite extensive research since the community learned about adversarial examples 10 years ago, we still do not know how to train high-accuracy classifiers that are guaranteed to be robust to small perturbations of their inputs. Previous works often argued that this might be because no classifier exists that is robust and accurate at the same time. However, in computer vision this assumption does not match reality where humans are usually accurate and robust on most tasks of interest. We offer an alternative explanation and show that in certain settings robust generalization is only possible with unrealistically large amounts of data. Specifically, we find a setting where a robust classifier exists, it is easy to learn an accurate classifier, yet it requires an exponential amount of data to learn a robust classifier. Based on this theoretical result, we evaluate the influence of the amount of training data on datasets such as CIFAR-10. Our findings indicate that the amount of training data is the main factor determining the robust performance. Furthermore we show that there are low magnitude directions in the data which are useful for non-robust generalization but are not available for robust classifiers. We provide code at https://github.com/berndprach/IntriguingProperties.
format Preprint
id arxiv_https___arxiv_org_abs_2412_04245
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Intriguing Properties of Robust Classification
Prach, Bernd
Lampert, Christoph H.
Computer Vision and Pattern Recognition
Despite extensive research since the community learned about adversarial examples 10 years ago, we still do not know how to train high-accuracy classifiers that are guaranteed to be robust to small perturbations of their inputs. Previous works often argued that this might be because no classifier exists that is robust and accurate at the same time. However, in computer vision this assumption does not match reality where humans are usually accurate and robust on most tasks of interest. We offer an alternative explanation and show that in certain settings robust generalization is only possible with unrealistically large amounts of data. Specifically, we find a setting where a robust classifier exists, it is easy to learn an accurate classifier, yet it requires an exponential amount of data to learn a robust classifier. Based on this theoretical result, we evaluate the influence of the amount of training data on datasets such as CIFAR-10. Our findings indicate that the amount of training data is the main factor determining the robust performance. Furthermore we show that there are low magnitude directions in the data which are useful for non-robust generalization but are not available for robust classifiers. We provide code at https://github.com/berndprach/IntriguingProperties.
title Intriguing Properties of Robust Classification
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2412.04245