PulseAugur
EN
LIVE 12:07:01

New benchmarks and frameworks advance AI model robustness evaluation

Researchers have introduced PRBench, a new benchmark designed to standardize the evaluation of probabilistic robustness in deep learning models. This benchmark compares various adversarial training (AT) and probabilistic robustness (PR) targeted training methods across multiple metrics including accuracy, robustness, training efficiency, and generalization error. Findings suggest that AT methods are more versatile for improving both adversarial and probabilistic robustness, while PR-targeted methods offer better generalization and clean accuracy. Separately, a new framework using the discrete modulus of continuity (DMOC) offers a data-driven approach to assess neural network robustness, moving beyond traditional Lipschitz continuity measures and proving effective on large datasets like ImageNet. AI

IMPACT New benchmarks and data-driven frameworks are emerging to better assess and improve the reliability of AI models against various perturbations.

RANK_REASON The cluster contains multiple academic papers introducing new benchmarks and methodologies for evaluating AI model robustness.

Read on arXiv cs.LG →

AI-generated summary · Google Gemini · from 4 sources. How we write summaries →

COVERAGE [4]

  1. arXiv cs.LG TIER_1 English(EN) · Yi Zhang, Zheng Wang, Zhen Chen, Wenjie Ruan, Qing Guo, Siddartha Khastgir, Carsten Maple, Xingyu Zhao ·

    PRBench: A Standardized Probabilistic Robustness Benchmark

    arXiv:2511.01724v3 Announce Type: replace-cross Abstract: 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…

  2. arXiv stat.ML TIER_1 English(EN) · J\"urgen D\"olz, Michael Multerer, Michele Palma ·

    Beyond Lipschitz: Data-Driven Robustness via Discrete Modulus of Continuity

    arXiv:2605.28729v1 Announce Type: new Abstract: Robustness of neural networks is commonly quantified via local or global Lipschitz constants. However, Lipschitz continuity can be overly coarse or overly restrictive as global robustness measure, failing to capture nuanced, data-de…

  3. arXiv stat.ML TIER_1 Français(FR) · Lorenzo Testa, Francesca Chiaromonte, Kathryn Roeder ·

    Rescuing double robustness: safe estimation under complete misspecification

    arXiv:2509.22446v2 Announce Type: replace-cross Abstract: Double robustness is a major selling point of semiparametric and missing data methodology. Its virtues lie in protection against partial nuisance misspecification and asymptotic semiparametric efficiency under correct nuis…

  4. arXiv stat.ML TIER_1 English(EN) · Michele Palma ·

    Beyond Lipschitz: Data-Driven Robustness via Discrete Modulus of Continuity

    Robustness of neural networks is commonly quantified via local or global Lipschitz constants. However, Lipschitz continuity can be overly coarse or overly restrictive as global robustness measure, failing to capture nuanced, data-dependent behavior. We propose a data-driven, arch…