AI Paper Atlas

Research map / Large language models / Alignment and preferences

Interpretable Reward Model via Sparse Autoencoder

Shuyi Zhang, Wei Shi, Sihang Li, Jiayi Liao, Tao Liang, Hengxing Cai, Xiang Wang

AAAI 2026 · Large language models › Alignment and preferences · approach: Reward modeling for diverse preferences · 12 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 pretrained SAE encoder instead of random init (ablation)SARM-4B · RewardBench 2Overall score: 68.4 -> 73.6Improved
pretrain SAE on sequence-level instead of token-level activations (ablation)SARM-4B · RewardBench 2Overall score: 71.5 -> 73.6Improved
pretrain SAE on sequence-level instead of token-level activations (ablation)SARM-4B · RewardBench 2Ties: 72.5 -> 79.6Improved
pretrain SAE on sequence-level instead of token-level activations (ablation)SARM-4B · RewardBench 2Safety: 92.9 -> 91.3Hurt
move SAE from layer 7 to layer 14 (ablation)SARM-2B · RewardBench 2Overall score: 36.21 -> 62.52Improved
move SAE from layer 14 to layer 21 (ablation)SARM-2B · RewardBench 2Overall score: 62.52 -> 64.17Improved
move SAE from layer 21 to layer 28 (ablation)SARM-2B · RewardBench 2Overall score: 64.17 -> 62.94Hurt
increase SAE feature dimension from 8x to 32x (ablation)SARM-2B · RewardBench 2Overall score: 63.04 -> 62.94No clear effect
increase SAE sparsity k from 48 to 240 (ablation)SARM-2B · RewardBench 2Overall score: 61.99 -> 62.89No clear effect
scale backbone from SARM-3B to SARM-4BSARM · RewardBench 2Overall score: 64.2 -> 73.6Improved
multiply value-head weight of a safety feature by a constantSARM · RewardBench 2reward distributionImproved

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