AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

A Provable Energy-Guided Test-Time Defense Boosting Adversarial Robustness of Large Vision-Language Models

Mujtaba Hussain Mirza, Antonio D'Orazio, Odelia Melamed, Iacopo Masi

CVPR 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Vision-language model robustness · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
apply ET3 test-time energy minimization to robust CLIP zero-shot classifiersTeCoA/FARE ViT-L/14 · 14-dataset zero-shot, AutoAttack eps 4/255robust accuracy: 32.85 -> 40.93 (TeCoA, eps_t 4/255)Improved
apply ET3 to models adversarially trained with the weaker eps_t 2/255FARE ViT-L/14 · 14-dataset zero-shot avgrobust accuracy: 20.39 -> 31.75Improved
apply ET3 to robust CLIPViT-L/14 TeCoA/FARE · 14-dataset zero-shot avgclean accuracy: 67.57 -> 67.42 (FARE eps_t 2/255)No clear effect
add ET3 to the lightweight TTC-style transformation baselineTeCoA CLIP-ViT-B/32 · 8 fine-grained datasets, PGD-100robust accuracy: 13.90 (TTC) -> 16.51 (ET3)Improved
combine ET3 with augmentation ensembleEnsemble · 8 fine-grained datasets, PGD-100robust accuracy: 19.4 -> 22.26Improved
combine ET3 with R-TPTR-TPT · 8 fine-grained datasets, PGD-100robust accuracy: 21.41 -> 23.88Improved
apply ET3 to LLaVA vision encoder, standard CLIPLLaVA 1.5-7B · captioning and VQA avg, eps 4/255robust score: 1.0 -> 38.8Improved
apply ET3 to LLaVA with standard CLIPLLaVA 1.5-7B · captioning and VQA avg, cleanclean score: 76.2 -> 73.9Hurt
apply ET3 under defense-aware adaptive attack on ImageNet RN-50Salman et al. RN-50 · ImageNet worst-case robustnessrobust accuracy: 34.96 -> 37.70Improved
apply ET3 to RN-50 on clean ImageNetSalman et al. RN-50 · ImageNet cleanclean accuracy: 64.02 -> 63.12Hurt
apply prior TTT defense Singh et al. to RN-50Salman et al. RN-50 · ImageNet worst-case robustnessrobust accuracy: 34.96 -> 34.68No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy