Adversarial Robustness of Sparse Local Lipschitz Predictors
SIAM Journal on Mathematics of Data Science · 31.12.2023 · 5 (4) · 920-948
DOI: 10.1137/22m1478835 ↗Kurz gefasst
Fachartikel in der Zeitschrift SIAM Journal on Mathematics of Data Science, online erschienen am 30.10.2023. Verfasst von Ramchandran Muthukumar und 1 weiteren Autor:innen. Die Arbeit wurde bisher 11-mal zitiert (Crossref) und nennt 65 Literaturangaben. Gefördert von: Defense Advanced Research Projects Agency, National Science Foundation.
Abstract (Originalsprache)
This work studies the adversarial robustness of parametric functions composed of a linear predictor and a nonlinear representation map. Our analysis relies on sparse local Lipschitzness (SLL), an extension of local Lipschitz continuity that better captures the stability and reduced effective dimensionality of predictors upon local perturbations. SLL functions preserve a certain degree of structure, given by the sparsity pattern in the representation map, and include several popular hypothesis classes, such as piecewise linear models, Lasso and its variants, and deep feedforward ReLU networks. Compared with traditional Lipschitz analysis, we provide a tighter robustness certificate on the minimal energy of an adversarial example, as well as tighter data-dependent nonuniform bounds on the robust generalization error of these predictors. We instantiate these results for the case of deep neural networks and provide numerical evidence that supports our results, shedding new insights into natural regularization strategies to increase the robustness of these models.
Förderung
- Defense Advanced Research Projects Agency
- National Science Foundation
Zitieren
APA
Muthukumar, R., & Sulam, J. (2023). Adversarial Robustness of Sparse Local Lipschitz Predictors. SIAM Journal on Mathematics of Data Science, 5(4), 920-948. https://doi.org/10.1137/22m1478835
Vancouver
Muthukumar R, Sulam J. Adversarial Robustness of Sparse Local Lipschitz Predictors. SIAM Journal on Mathematics of Data Science. 2023;5(4):920-948. doi:10.1137/22m1478835
BibTeX
@article{Muthukumar2023,
author = {Muthukumar, Ramchandran and Sulam, Jeremias},
title = {Adversarial Robustness of Sparse Local Lipschitz Predictors},
journal = {SIAM Journal on Mathematics of Data Science},
year = {2023},
volume = {5},
number = {4},
pages = {920--948},
doi = {10.1137/22m1478835}
}
EUVIMED