Saved in:
Bibliographic Details
Main Authors: Zhang, Yi, Wang, Zheng, Chen, Zhen, Ruan, Wenjie, Guo, Qing, Khastgir, Siddartha, Maple, Carsten, Zhao, Xingyu
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2511.01724
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917535261130752
author Zhang, Yi
Wang, Zheng
Chen, Zhen
Ruan, Wenjie
Guo, Qing
Khastgir, Siddartha
Maple, Carsten
Zhao, Xingyu
author_facet Zhang, Yi
Wang, Zheng
Chen, Zhen
Ruan, Wenjie
Guo, Qing
Khastgir, Siddartha
Maple, Carsten
Zhao, Xingyu
contents Deep learning models are notoriously vulnerable to imperceptible perturbations. Most existing research centers on adversarial robustness (AR), which evaluates models under worst-case scenarios by examining the existence of deterministic adversarial examples (AEs). In contrast, probabilistic robustness (PR) adopts a statistical perspective, measuring the probability that predictions remain correct under stochastic perturbations. While PR is widely regarded as a practical complement to AR, dedicated training methods for improving PR are still relatively underexplored, albeit with emerging progress. Among the few PR-targeted training methods, we identify three limitations: i non-comparable evaluation protocols; ii limited comparisons to strong AT baselines despite anecdotal PR gains from AT; and iii no unified framework to compare the generalization of these methods. Thus, we introduce PRBench, the first benchmark dedicated to evaluating improvements in PR achieved by different robustness training methods. PRBench empirically compares most common AT and PR-targeted training methods using a comprehensive set of metrics, including clean accuracy, PR and AR performance, training efficiency, and generalization error (GE). We also provide theoretical analysis on the GE of PR performance across different training methods. Main findings revealed by PRBench include: AT methods are more versatile than PR-targeted training methods in terms of improving both AR and PR performance across diverse hyperparameter settings, while PR-targeted training methods consistently yield lower GE and higher clean accuracy. A leaderboard comprising 229 trained models across 7 datasets and 10 model architectures is publicly available at https://wellzline.github.io/PRBenchLeaderboard/.
format Preprint
id arxiv_https___arxiv_org_abs_2511_01724
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle PRBench: A Standardized Probabilistic Robustness Benchmark
Zhang, Yi
Wang, Zheng
Chen, Zhen
Ruan, Wenjie
Guo, Qing
Khastgir, Siddartha
Maple, Carsten
Zhao, Xingyu
Computer Vision and Pattern Recognition
Machine Learning
Deep learning models are notoriously vulnerable to imperceptible perturbations. Most existing research centers on adversarial robustness (AR), which evaluates models under worst-case scenarios by examining the existence of deterministic adversarial examples (AEs). In contrast, probabilistic robustness (PR) adopts a statistical perspective, measuring the probability that predictions remain correct under stochastic perturbations. While PR is widely regarded as a practical complement to AR, dedicated training methods for improving PR are still relatively underexplored, albeit with emerging progress. Among the few PR-targeted training methods, we identify three limitations: i non-comparable evaluation protocols; ii limited comparisons to strong AT baselines despite anecdotal PR gains from AT; and iii no unified framework to compare the generalization of these methods. Thus, we introduce PRBench, the first benchmark dedicated to evaluating improvements in PR achieved by different robustness training methods. PRBench empirically compares most common AT and PR-targeted training methods using a comprehensive set of metrics, including clean accuracy, PR and AR performance, training efficiency, and generalization error (GE). We also provide theoretical analysis on the GE of PR performance across different training methods. Main findings revealed by PRBench include: AT methods are more versatile than PR-targeted training methods in terms of improving both AR and PR performance across diverse hyperparameter settings, while PR-targeted training methods consistently yield lower GE and higher clean accuracy. A leaderboard comprising 229 trained models across 7 datasets and 10 model architectures is publicly available at https://wellzline.github.io/PRBenchLeaderboard/.
title PRBench: A Standardized Probabilistic Robustness Benchmark
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2511.01724