Certified vs. Empirical Adversarial Robust-ness via Hybrid Convolutions with Attention Stochasticity

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Dhar, Joy, Xia, Song, Pandey, Manish Kumar, Haghighat, Maryam, Alavi, Azadeh, Sohel, Ferdous, Zhang, Wenyu, Zaidi, Nayyar
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911640571609088
author Dhar, Joy
Xia, Song
Pandey, Manish Kumar
Haghighat, Maryam
Alavi, Azadeh
Sohel, Ferdous
Zhang, Wenyu
Zaidi, Nayyar
author_facet Dhar, Joy
Xia, Song
Pandey, Manish Kumar
Haghighat, Maryam
Alavi, Azadeh
Sohel, Ferdous
Zhang, Wenyu
Zaidi, Nayyar
contents We introduce Hybrid Convolutions with Attention Stochasticity (HyCAS), an adversarial defense that narrows the long-standing gap between provable robustness under L2 certificates and empirical robustness against strong L attacks, while preserving strong generalization across diverse imaging benchmarks. HyCAS unifies deterministic and randomized principles by coupling 1-Lipschitz, spectrally normalized convolutions with two stochastic components, spectral normalized random, projection filters and a randomized attention-noise mechanism, to realize a randomized defense. Injecting smoothing randomness inside the architecture yields an overall <= 2-Lipschitz network with formal certificates. Exten-sive experiments on diverse imaging benchmarks, including CIFAR-10/100, ImageNet-1k, NIH Chest X-ray, HAM10000, show that HyCAS surpasses prior leading certified and empirical defenses, boosting certified accuracy by up to 7.3% (on NIH Chest X-ray) and empirical robustness by up to 3.1% (on HAM10000), without sacrificing clean accuracy. These results show that a randomized Lipschitz constrained architecture can simultaneously improve both certified L2 and empirical L adversarial robustness, thereby supporting safer deployment of deep models in high-stakes applications. Code: https://github.com/misti1203/HyCAS
format Preprint
id arxiv_https___arxiv_org_abs_2605_01519
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Certified vs. Empirical Adversarial Robust-ness via Hybrid Convolutions with Attention Stochasticity
Dhar, Joy
Xia, Song
Pandey, Manish Kumar
Haghighat, Maryam
Alavi, Azadeh
Sohel, Ferdous
Zhang, Wenyu
Zaidi, Nayyar
Computer Vision and Pattern Recognition
We introduce Hybrid Convolutions with Attention Stochasticity (HyCAS), an adversarial defense that narrows the long-standing gap between provable robustness under L2 certificates and empirical robustness against strong L attacks, while preserving strong generalization across diverse imaging benchmarks. HyCAS unifies deterministic and randomized principles by coupling 1-Lipschitz, spectrally normalized convolutions with two stochastic components, spectral normalized random, projection filters and a randomized attention-noise mechanism, to realize a randomized defense. Injecting smoothing randomness inside the architecture yields an overall <= 2-Lipschitz network with formal certificates. Exten-sive experiments on diverse imaging benchmarks, including CIFAR-10/100, ImageNet-1k, NIH Chest X-ray, HAM10000, show that HyCAS surpasses prior leading certified and empirical defenses, boosting certified accuracy by up to 7.3% (on NIH Chest X-ray) and empirical robustness by up to 3.1% (on HAM10000), without sacrificing clean accuracy. These results show that a randomized Lipschitz constrained architecture can simultaneously improve both certified L2 and empirical L adversarial robustness, thereby supporting safer deployment of deep models in high-stakes applications. Code: https://github.com/misti1203/HyCAS
title Certified vs. Empirical Adversarial Robust-ness via Hybrid Convolutions with Attention Stochasticity
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2605.01519