AI Paper Atlas

Research map / Label-efficient and robust learning / Test-time adaptation

Get RICH or Die Scaling: Profitably Trading Inference Compute for Robustness

Tavish Malcolm McDonald, Bo Lei, Stanislav Fort, Bhavya Kailkhura, Brian R. Bartoldson

ICLR 2026 · Label-efficient and robust learning › Test-time adaptation · approach: Extrapolation and single-sample adaptation · 1 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
add explicit security specificationDelta2-LLaVA-v1.5 (high robustness) · Visual prompt injection PGD step 300attacker loss: 12.4 -> 21.1Improved
add explicit security specificationFARE-LLaVA-v1.5 (medium robustness) · Visual prompt injection PGD step 300attacker loss: 7.0 -> 7.2No clear effect
add explicit security specificationLLaVA-v1.5 (low robustness) · Visual prompt injection PGD step 100attacker loss: 6.4 -> 2.9Hurt
scale spec repetition from K=0 to K=5Delta2-LLaVA-v1.5 · PGD steps to attack success, eps=64/255PGD steps: 25.4 -> 63.2Improved
scale spec repetition from K=0 to K=5FARE-LLaVA-v1.5 · PGD steps to attack success, eps=64/255PGD steps: 6.7 -> 9.2Mixed
scale spec repetition from K=0 to K=5LLaVA-v1.5 · PGD steps to attack success, eps=64/255PGD steps: 6.2 -> 7.4No clear effect
shrink attack budget from 64/255 to 16/255 and scale K=0 to K=5 (ablation)FARE-LLaVA-v1.5 · PGD steps to attack successPGD steps: 18.8 -> 27.2Improved
add CoTLLaVA-v1.5 (non-robust) · Attack-Bard adversarial (30-way)accuracy: 38.0 -> 44.5No clear effect
add CoTFARE-LLaVA-v1.5 · Attack-Bard adversarial (30-way)accuracy: 56.0 -> 65.5Improved
add CoTDelta2-LLaVA-v1.5 · Attack-Bard adversarial (30-way)accuracy: 62.0 -> 73.0Improved
add CoTQwen-2.5-VL-72B · Attack-Bard adversarial (1000-way)accuracy: 13.0 -> 18.0No clear effect
add CoTLlama-3.2-Vision-90B · Attack-Bard adversarial (1000-way)accuracy: 27.0 -> 27.5No clear effect

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