AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

CorrSteer: Generation-Time LLM Steering via Correlated Sparse Autoencoder Features

Seonglae Cho, Zekun Wu, Adriano Koshiyama

ICML 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Low-rank compression and merging · 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
select SAE features by generation-time correlation and steer (CorrSteer-A)Gemma-2 2B · MMLUaccuracy: 52.21 -> 55.48Improved
add validation-based pruning of features (CorrSteer-P over CorrSteer-A) (ablation)CorrSteer · BBQ Ambigaccuracy: 62.06 -> 66.00Improved
use many layer-wise features (CorrSteer-A) instead of a single feature (CorrSteer-S) (ablation)CorrSteer · Eight benchmarks, Gemma-2 2Baccuracy: A beats S on 5 of 8 tasksMixed
use max-pooling instead of mean-pooling for feature aggregation (ablation)CorrSteer · HarmBenchrefusal rate: 0.00 -> 67.50Improved
use max-pooling instead of all-token pooling (ablation)CorrSteer · MMLUaccuracy: 52.91 -> 56.32Improved
steer with positively correlated features only, versus negatively correlated features across layers (ablation)CorrSteer · BBQ Disambigaccuracy: 12.15 -> 76.53Improved
negatively correlated single feature (Neg-S) steering (ablation)CorrSteer · MMLUaccuracy: 52.23 -> 52.24No clear effect
correlation-based selection instead of random feature selection (ablation)CorrSteer-A · MMLUaccuracy: 6.29 -> 55.48Improved
use true correctness labels instead of permuted labels (ablation)CorrSteer-A · MMLUaccuracy: 6.24 -> 55.48Improved
keep only semantic features, remove structural/formatting features (ablation)CorrSteer-A · BBQ Ambigaccuracy: 62.06 -> 63.93Improved
keep only semantic features on MMLU (ablation)CorrSteer-A · MMLUaccuracy: 55.48 -> 55.12No clear effect
scale steering coefficient from 1.0x to 2.0x (ablation)CorrSteer on HarmBench · HarmBench, Gemma-2 2B seed 42refusal rate: 60.36% -> 7.50%Hurt

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