Interpolated Adversarial Training: Achieving robust neural networks without sacrificing too much accuracy
dc.contributor | Aalto-yliopisto | fi |
dc.contributor | Aalto University | en |
dc.contributor.author | Lamb, Alex | en_US |
dc.contributor.author | Verma, Vikas | en_US |
dc.contributor.author | Kawaguchi, Kenji | en_US |
dc.contributor.author | Matyasko, Alexander | en_US |
dc.contributor.author | Khosla, Savya | en_US |
dc.contributor.author | Kannala, Juho | en_US |
dc.contributor.author | Bengio, Yoshua | en_US |
dc.contributor.department | Department of Computer Science | en |
dc.contributor.groupauthor | Professorship Kannala Juho | en |
dc.contributor.groupauthor | Computer Science Professors | en |
dc.contributor.groupauthor | Computer Science - Visual Computing (VisualComputing) | en |
dc.contributor.groupauthor | Computer Science - Artificial Intelligence and Machine Learning (AIML) | en |
dc.contributor.organization | Université de Montréal | en_US |
dc.contributor.organization | Harvard University | en_US |
dc.contributor.organization | National University of Singapore Museum | en_US |
dc.contributor.organization | Google, India | en_US |
dc.date.accessioned | 2023-01-18T09:20:48Z | |
dc.date.available | 2023-01-18T09:20:48Z | |
dc.date.issued | 2022-10 | en_US |
dc.description.abstract | Adversarial 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.version | Peer reviewed | en |
dc.format.extent | 16 | |
dc.format.extent | 218-233 | |
dc.format.mimetype | application/pdf | en_US |
dc.identifier.citation | Lamb, 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 . https://doi.org/10.1016/j.neunet.2022.07.012 | en |
dc.identifier.doi | 10.1016/j.neunet.2022.07.012 | en_US |
dc.identifier.issn | 0893-6080 | |
dc.identifier.issn | 1879-2782 | |
dc.identifier.other | PURE UUID: 328816d1-cb56-4720-90ed-9299e237f163 | en_US |
dc.identifier.other | PURE ITEMURL: https://research.aalto.fi/en/publications/328816d1-cb56-4720-90ed-9299e237f163 | en_US |
dc.identifier.other | PURE LINK: http://www.scopus.com/inward/record.url?scp=85135390093&partnerID=8YFLogxK | en_US |
dc.identifier.other | PURE FILEURL: https://research.aalto.fi/files/97683506/Interpolated_Adversarial_Training.pdf | en_US |
dc.identifier.uri | https://aaltodoc.aalto.fi/handle/123456789/118819 | |
dc.identifier.urn | URN:NBN:fi:aalto-202301181175 | |
dc.language.iso | en | en |
dc.publisher | Elsevier Ltd | |
dc.relation.ispartofseries | Neural Networks | en |
dc.relation.ispartofseries | Volume 154 | en |
dc.rights | openAccess | en |
dc.subject.keyword | Adversarial robustness | en_US |
dc.subject.keyword | Mixup | en_US |
dc.subject.keyword | Manifold Mixup | en_US |
dc.subject.keyword | Standard test error | en_US |
dc.title | Interpolated Adversarial Training: Achieving robust neural networks without sacrificing too much accuracy | en |
dc.type | A1 Alkuperäisartikkeli tieteellisessä aikakauslehdessä | fi |
dc.type.version | publishedVersion |