How to Compare Adversarial Robustness of Classifiers from a Global Perspective

Risse N, Göpfert C, Göpfert JP (2021)
In: Artificial Neural Networks and Machine Learning – ICANN 2021. 30th International Conference on Artificial Neural Networks, Bratislava, Slovakia, September 14–17, 2021, Proceedings, Part I. Farkaš I, Masulli P, Otte S, Wermter S (Eds); Lecture Notes in Computer Science, 12891. Cham: Springer International Publishing: 29-41.

Konferenzbeitrag | Veröffentlicht | Englisch
 
Download
Es wurden keine Dateien hochgeladen. Nur Publikationsnachweis!
Herausgeber*in
Farkaš, Igor; Masulli, Paolo; Otte, Sebastian; Wermter, Stefan
Abstract / Bemerkung
Adversarial robustness of machine learning models has attracted considerable attention over recent years. Adversarial attacks undermine the reliability of and trust in machine learning models, but the construction of more robust models hinges on a rigorous understanding of adversarial robustness as a property of a given model. Point-wise measures for specific threat models are currently the most popular tool for comparing the robustness of classifiers and are used in most recent publications on adversarial robustness. In this work, we use robustness curves to show that point-wise measures fail to capture important global properties that are essential to reliably compare the robustness of different classifiers. We introduce new ways in which robustness curves can be used to systematically uncover these properties and provide concrete recommendations for researchers and practitioners when assessing and comparing the robustness of trained models. Furthermore, we characterize scale as a way to distinguish small and large perturbations, and relate it to inherent properties of data sets, demonstrating that robustness thresholds must be chosen accordingly. We hope that our work contributes to a shift of focus away from point-wise measures of robustness and towards a discussion of the question what kind of robustness could and should reasonably be expected. We release code to reproduce all experiments presented in this paper, which includes a Python module to calculate robustness curves for arbitrary data sets and classifiers, supporting a number of frameworks, including TensorFlow, PyTorch and JAX.
Erscheinungsjahr
2021
Titel des Konferenzbandes
Artificial Neural Networks and Machine Learning – ICANN 2021. 30th International Conference on Artificial Neural Networks, Bratislava, Slovakia, September 14–17, 2021, Proceedings, Part I
forms.conference.field.series_title_volume.series_title.label
Lecture Notes in Computer Science
Band
12891
Seite(n)
29-41
Konferenz
30th International Conference on Artificial Neural Networks (ICANN 2021)
Konferenzort
Bratislava, Slovakia
Konferenzdatum
2021-09-14 – 2021-09-17
ISBN
978-3-030-86361-6
eISBN
978-3-030-86362-3
Page URI
https://pub.uni-bielefeld.de/record/2957385

Zitieren

Risse N, Göpfert C, Göpfert JP. How to Compare Adversarial Robustness of Classifiers from a Global Perspective. In: Farkaš I, Masulli P, Otte S, Wermter S, eds. Artificial Neural Networks and Machine Learning – ICANN 2021. 30th International Conference on Artificial Neural Networks, Bratislava, Slovakia, September 14–17, 2021, Proceedings, Part I. Lecture Notes in Computer Science. Vol 12891. Cham: Springer International Publishing; 2021: 29-41.
Risse, N., Göpfert, C., & Göpfert, J. P. (2021). How to Compare Adversarial Robustness of Classifiers from a Global Perspective. In I. Farkaš, P. Masulli, S. Otte, & S. Wermter (Eds.), Lecture Notes in Computer Science: Vol. 12891. Artificial Neural Networks and Machine Learning – ICANN 2021. 30th International Conference on Artificial Neural Networks, Bratislava, Slovakia, September 14–17, 2021, Proceedings, Part I (pp. 29-41). Cham: Springer International Publishing. https://doi.org/10.1007/978-3-030-86362-3_3
Risse, N., Göpfert, C., and Göpfert, J. P. (2021). “How to Compare Adversarial Robustness of Classifiers from a Global Perspective” in Artificial Neural Networks and Machine Learning – ICANN 2021. 30th International Conference on Artificial Neural Networks, Bratislava, Slovakia, September 14–17, 2021, Proceedings, Part I, Farkaš, I., Masulli, P., Otte, S., and Wermter, S. eds. Lecture Notes in Computer Science, vol. 12891, (Cham: Springer International Publishing), 29-41.
Risse, N., Göpfert, C., & Göpfert, J.P., 2021. How to Compare Adversarial Robustness of Classifiers from a Global Perspective. In I. Farkaš, et al., eds. Artificial Neural Networks and Machine Learning – ICANN 2021. 30th International Conference on Artificial Neural Networks, Bratislava, Slovakia, September 14–17, 2021, Proceedings, Part I. Lecture Notes in Computer Science. no.12891 Cham: Springer International Publishing, pp. 29-41.
N. Risse, C. Göpfert, and J.P. Göpfert, “How to Compare Adversarial Robustness of Classifiers from a Global Perspective”, Artificial Neural Networks and Machine Learning – ICANN 2021. 30th International Conference on Artificial Neural Networks, Bratislava, Slovakia, September 14–17, 2021, Proceedings, Part I, I. Farkaš, et al., eds., Lecture Notes in Computer Science, vol. 12891, Cham: Springer International Publishing, 2021, pp.29-41.
Risse, N., Göpfert, C., Göpfert, J.P.: How to Compare Adversarial Robustness of Classifiers from a Global Perspective. In: Farkaš, I., Masulli, P., Otte, S., and Wermter, S. (eds.) Artificial Neural Networks and Machine Learning – ICANN 2021. 30th International Conference on Artificial Neural Networks, Bratislava, Slovakia, September 14–17, 2021, Proceedings, Part I. Lecture Notes in Computer Science. 12891, p. 29-41. Springer International Publishing, Cham (2021).
Risse, Niklas, Göpfert, Christina, and Göpfert, Jan Philip. “How to Compare Adversarial Robustness of Classifiers from a Global Perspective”. Artificial Neural Networks and Machine Learning – ICANN 2021. 30th International Conference on Artificial Neural Networks, Bratislava, Slovakia, September 14–17, 2021, Proceedings, Part I. Ed. Igor Farkaš, Paolo Masulli, Sebastian Otte, and Stefan Wermter. Cham: Springer International Publishing, 2021.Vol. 12891. Lecture Notes in Computer Science. 29-41.

Export

Markieren/ Markierung löschen
Markierte Publikationen

Open Data PUB

Suchen in

Google Scholar
ISBN Suche