Get Started
Home
Topics
Search
Library
Research questionHow can one vision model remain robust across changing and unseen adversarial perturbation budgets?Adversarially trained classifiers often specialize in a single perturbation budget. Deployments facing changing attack strengths may therefore require multiple models, making broad threat coverage costly.
Alignment & Safety
Computer Vision
Image & Video Processing
Machine Learning
Statistical Machine Learning
Latest papersRecent research connected to this question, newest first.Towards One-for-All Robustness Across a Continuum of Threat LevelsThe source studies image classification on CIFAR-10, CIFAR-100, and Tiny-ImageNet using a model conditioned on perturbation level and trained over a distribution of budgets. It reports generalization to unseen budgets and transfer under mismatched threat conditions, with 4.6% parameter overhead; evidence is limited to these experimental settings.research paper · Sep 2, 2026
Related questions
How can image-poisoning attacks remain effective across supervised and self-supervised learning without obvious visual artifacts?How can deployed vision models resist diverse backdoor triggers without model internals or auxiliary data?How can frozen dense-prediction models remain robust under distribution shift without retraining or changing their prediction heads?Can small targeted grayscale patches force chosen semantics in infrared vision-language models across tasks?