AI Paper Atlas

Research map / Large language models / Alignment and preferences

Factored Causal Representation Learning for Robust Reward Modeling in RLHF

Yupei Yang, Lin Yang, Wanxi Deng, Lin Qu, Fan Feng, Biwei Huang, Shikui Tu, Lei Xu

ICML 2026 · Large language models › Alignment and preferences · approach: Reward modeling for diverse preferences · 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
factorize latent into causal and non-causal parts (vs single bottleneck latent, InfoRM-style) (ablation)CausalRM · Math reward model ID avgpairwise accuracy (%): 66.1 -> 70.1Improved
add latent factorization (ablation)CausalRM · Math reward model length sensitivitysigma_len: 0.14 -> 0.03Improved
add adversarial head with gradient reversal (ablation)CausalRM · Math reward model OOD avgpairwise accuracy (%): 81.5 -> 85.6Improved
add adversarial head with gradient reversal (ablation)CausalRM · Math length sensitivitysigma_len: 0.13 -> 0.03Improved
add reconstruction decoder loss (ablation)CausalRM · Math reward model OOD avgpairwise accuracy (%): 83.1 -> 85.6Improved
add KL regularization on causal latent zc (ablation)CausalRM · Math reward model OOD avgpairwise accuracy (%): 81.1 -> 85.6Improved
add KL regularization on non-causal latent znc (ablation)CausalRM · Math reward model ID avgpairwise accuracy (%): 64.0 -> 70.1Improved
add KL regularization on both latents (ablation)CausalRM · Sycophancy-hacked test ID avgpairwise accuracy (%): 46.0 -> 70.6Improved
replace Standard RM with CausalRM reward in PPOQwen2.5-Math-7B RLHF · Math RLHF ID avgfinal-answer accuracy (%): 67.8 -> 74.0Improved
use InfoRM reward in PPO instead of Standard RMQwen2.5-Math-7B RLHF · Math RLHF ID avgfinal-answer accuracy (%): 67.8 -> 48.0Hurt
use CausalRM instead of Standard RM under sycophantic prefix artifactreward model on hacked Anthropic-HH · Sycophancy-hacked ID avgpairwise accuracy (%): 59.2 -> 70.6Improved
use CausalRM instead of artifact-specific ODIN/MMD for length biasmath reward model · Math ID avgpairwise accuracy (%): 69.3 -> 70.1 (vs MMD); 68.8 -> 70.1 (vs ODIN)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