AI Paper Atlas

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

Self-Calibrated Consistency can Fight Back for Adversarial Robustness in Vision-Language Models

Jiaxiang Liu, Jiawei Du, Xiao Liu, Shangyang Li, Songchen Ma, Changshuo Wang, Prayag Tiwari, Mingkun Xu

ICML 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Vision-language model robustness

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
add semantic consistency (ablation)SCC without modules · 16-dataset avg, PGD-10 eps=1/255 (ablation)robust accuracy (%): 39.18 -> 39.76Improved
add spatial consistency (ablation)SCC without modules · 16-dataset avg, PGD-10 eps=1/255 (ablation)robust accuracy (%): 39.18 -> 48.01Improved
add spatial consistency on top of semantic consistency (ablation)SCC semantic only · 16-dataset avg, PGD-10 eps=1/255 (ablation)robust accuracy (%): 39.76 -> 51.68Improved
add semantic consistency on top of spatial consistency (ablation)SCC spatial only · 16-dataset avg, PGD-10 eps=1/255 (ablation)robust accuracy (%): 48.01 -> 51.68Improved
replace TTC with SCCTTC · 16-dataset avg, PGD-10 eps=1/255robust accuracy (%): 39.17 -> 51.68Improved
replace TTC with SCC at larger attack budgetTTC · 16-dataset avg, PGD-10 eps=4/255robust accuracy (%): 20.63 -> 27.88Improved
replace TTC with SCC under CW attackTTC · 16-dataset avg, CW-10 eps=1/255robust accuracy (%): 38.17 -> 49.42Improved
replace TTC with SCC on BioMedCLIPTTC · 6 medical datasets avg, eps=1/255robust accuracy (%): 10.62 -> 34.00Improved
replace TTC with SCCTTC · DTD, adversarial attackrobust accuracy (%): 27.4 -> 34.6Improved
apply SCC instead of vanilla CLIPCLIP · 16-dataset avg, PGD-10 eps=1/255clean accuracy (%): 61.51 -> 60.21Hurt
use clean-image predictions for pseudo-labels instead of warm-up pseudo-labelsSCC · 16-dataset avg, PGD-10 eps=1/255robust accuracy (%): 51.68 -> 53.62Improved
increase number of views L from 1 to 2-4 (ablation)SCC · 16-dataset avg (sensitivity)robust accuracy (%)Improved

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