It begins with a boundary: A geometric view on probabilistically robust learning
ESAIM: Control, Optimisation and Calculus of Variations · 2026 · 32 · 56
DOI: 10.1051/cocv/2026040 ↗Kurz gefasst
Fachartikel in der Zeitschrift ESAIM: Control, Optimisation and Calculus of Variations, online erschienen am 07.07.2026. Verfasst von Leon Bungert und 5 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Division of Mathematical Sciences.
Abstract (Originalsprache)
Although deep neural networks have achieved super-human performance on many classification tasks, they often exhibit a worrying lack of robustness towards adversarially generated examples. Thus, considerable effort has been invested into reformulating standard Risk Minimization (RM) into an adversarially robust framework. Recently, attention has shifted towards approaches which interpolate between the robustness offered by adversarial training and the higher clean accuracy and faster training times of RM. In this paper, we take a fresh and geometric view on one such method – Probabilistically Robust Learning (PRL) [A. Robey et al. International Conference on Machine Learning. PMLR (2022) 18667–18686]. We propose a mathematical framework for understanding PRL, which allows us to identify geometric pathologies in its original formulation and to introduce a family of probabilistic nonlocal perimeter functionals to rectify them. We prove existence of solutions to the original and modified problems using relaxation methods in the calculus of variations and also study properties, as well as local limits, of the introduced perimeters. We also clarify, through a suitable Γ-convergence analysis, the way in which the original and modified PRL models interpolate between risk minimization and adversarial training.
Förderung
- Division of Mathematical Sciences
Zitieren
APA
Bungert, L., Trillos, N. G., Jacobs, M., McKenzie, D., Nikolić, Đ., & Wang, Q. (2026). It begins with a boundary: A geometric view on probabilistically robust learning. ESAIM: Control, Optimisation and Calculus of Variations, 32, 56. https://doi.org/10.1051/cocv/2026040
Vancouver
Bungert L, Trillos NG, Jacobs M, McKenzie D, Nikolić Đ, Wang Q. It begins with a boundary: A geometric view on probabilistically robust learning. ESAIM: Control, Optimisation and Calculus of Variations. 2026;32:56. doi:10.1051/cocv/2026040
BibTeX
@article{Bungert2026,
author = {Bungert, Leon and Trillos, Nicolás García and Jacobs, Matt and McKenzie, Daniel and Nikolić, Đorde and Wang, Qingsong},
title = {It begins with a boundary: A geometric view on probabilistically robust learning},
journal = {ESAIM: Control, Optimisation and Calculus of Variations},
year = {2026},
volume = {32},
pages = {56},
doi = {10.1051/cocv/2026040}
}
EUVIMED