Interpolated Adversarial Training: Achieving robust neural networks without sacrificing too much accuracy

dc.contributorAalto Universityen
dc.contributor.authorLamb, Alexen_US
dc.contributor.authorVerma, Vikasen_US
dc.contributor.authorKawaguchi, Kenjien_US
dc.contributor.authorMatyasko, Alexanderen_US
dc.contributor.authorKhosla, Savyaen_US
dc.contributor.authorKannala, Juhoen_US
dc.contributor.authorBengio, Yoshuaen_US
dc.contributor.departmentDepartment of Computer Scienceen
dc.contributor.groupauthorProfessorship Kannala Juhoen
dc.contributor.groupauthorComputer Science Professorsen
dc.contributor.groupauthorComputer Science - Visual Computing (VisualComputing)en
dc.contributor.groupauthorComputer Science - Artificial Intelligence and Machine Learning (AIML)en
dc.contributor.organizationUniversité de Montréalen_US
dc.contributor.organizationHarvard Universityen_US
dc.contributor.organizationNational University of Singapore Museumen_US
dc.contributor.organizationGoogle, Indiaen_US
dc.description.abstractAdversarial robustness has become a central goal in deep learning, both in the theory and the practice. However, successful methods to improve the adversarial robustness (such as adversarial training) greatly hurt generalization performance on the unperturbed data. This could have a major impact on how the adversarial robustness affects real world systems (i.e. many may opt to forego robustness if it can improve accuracy on the unperturbed data). We propose Interpolated Adversarial Training, which employs recently proposed interpolation based training methods in the framework of adversarial training. On CIFAR-10, adversarial training increases the standard test error ( when there is no adversary) from 4.43% to 12.32%, whereas with our Interpolated adversarial training we retain the adversarial robustness while achieving a standard test error of only 6.45%. With our technique, the relative increase in the standard error for the robust model is reduced from 178.1% to just 45.5%. Moreover, we provide mathematical analysis of Interpolated Adversarial Training to confirm its efficiencies and demonstrate its advantages in terms of robustness and generalization. (C) 2022 The Authors. Published by Elsevier Ltd.en
dc.description.versionPeer revieweden
dc.identifier.citationLamb, A, Verma, V, Kawaguchi, K, Matyasko, A, Khosla, S, Kannala, J & Bengio, Y 2022, ' Interpolated Adversarial Training: Achieving robust neural networks without sacrificing too much accuracy ', Neural Networks, vol. 154, pp. 218-233 .
dc.identifier.otherPURE UUID: 328816d1-cb56-4720-90ed-9299e237f163en_US
dc.identifier.otherPURE ITEMURL:
dc.identifier.otherPURE LINK:
dc.identifier.otherPURE FILEURL:
dc.publisherElsevier Ltd
dc.relation.ispartofseriesNeural Networksen
dc.relation.ispartofseriesVolume 154en
dc.subject.keywordAdversarial robustnessen_US
dc.subject.keywordManifold Mixupen_US
dc.subject.keywordStandard test erroren_US
dc.titleInterpolated Adversarial Training: Achieving robust neural networks without sacrificing too much accuracyen
dc.typeA1 Alkuperäisartikkeli tieteellisessä aikakauslehdessäfi