AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

Concept Heterogeneity-aware Representation Steering

Laziz Abdullaev, Noelle Y. L. Wong, Ryan Lee T. Z., Shiqi Jiang, Minh-Khoi Nguyen-Nhat, Tan Minh Nguyen

ICML 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Probabilistic embeddings and relational inference · 0 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
replace ActAdd difference-in-means with CHaRSActAdd · AdvBench jailbreak, Gemma2-9B-InstructASR: 91.35 -> 98.08Improved
replace ActAdd difference-in-means with CHaRSActAdd · AdvBench jailbreak, Qwen2.5-32B-InstructASR: 82.69 -> 89.42Improved
replace DirAbl difference-in-means with CHaRSDirAbl · AdvBench jailbreak, Gemma2-9B-InstructASR: 79.81 -> 84.62Improved
replace DirAbl difference-in-means with CHaRS-PCTDirAbl · AdvBench jailbreak, Gemma2-9B-InstructASR: 79.81 -> 85.58Improved
replace DirAbl difference-in-means with CHaRSDirAbl · AdvBench jailbreak, Qwen2.5-14B-InstructASR: 78.85 -> 80.77Improved
add principal component thresholding to CHaRSCHaRS with ActAdd · AdvBench jailbreak, Llama3.1-8B-InstructASR: 98.08 -> 99.04Improved
add principal component thresholding to CHaRSCHaRS with ActAdd · AdvBench jailbreak, Qwen2.5-7B-InstructASR: 95.19 -> 93.27Hurt
steer with ActAdd instead of no steeringunsteered Gemma2-9B-Instruct · tinyBenchmarks averageaccuracy: 71.11 -> 51.39Hurt
steer with DirAbl instead of no steeringunsteered Gemma2-9B-Instruct · tinyBenchmarks averageaccuracy: 71.11 -> 69.87No clear effect
use OT coupling instead of nearest-neighbor coupling for cluster matching (ablation)CHaRS ActAdd · AdvBench jailbreak, Qwen2.5-7B-Instruct, K=5ASR: 76.92 -> 95.19Improved
use OT coupling instead of uniform coupling for cluster matching (ablation)CHaRS ActAdd · AdvBench jailbreak, Qwen2.5-7B-Instruct, K=5ASR: 42.31 -> 95.19Improved
use more than one cluster (K>1) instead of K=1 (ablation)CHaRS ActAdd · AdvBench jailbreak, Qwen2.5-7B-InstructASR: 91.35 -> 95.19Improved

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