AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

REAL: Regression-Aware Reinforcement Learning for LLM-as-a-Judge

Yasi Zhang, Tianyu Chen, Mingyuan Zhou, Oscar Leong, Ying Nian Wu, Michal Lukasik

ICML 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning · 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 binary-accuracy RL with REAL regression-aware policy-dependent rewardRLOO RL on Mistral2-7B · Judge benchmarks averagePearson: 64.1 -> 67.9Improved
use regression-aware but policy-independent reward (Standard RL(reg)) instead of accuracy reward (ablation)Standard RL on Mistral2-7B · Judge benchmarks averagePearson: 64.1 -> 63.4No clear effect
add RL with REAL on top of TRACT SFT checkpointTRACT on Mistral2-7B · Judge benchmarks averagePearson: 63.2 -> 67.9Improved
add RL with REAL on top of RAFT SFT checkpointRAFT on Qwen3-32B · Judge benchmarks averagePearson: 62.6 -> 71.0Improved
add RL with REAL on top of RAFT SFT checkpointRAFT on Qwen3-8B · Judge benchmarks averagePearson: 61.9 -> 67.0Improved
increase group size K from 4 to 16 (ablation)REAL · Judge benchmarks averagePearson: 63.7 -> 67.9Improved
increase group size K from 8 to 16 (ablation)REAL · Judge benchmarks averagePearson: 67.2 -> 67.9Improved
add log-likelihood term (lambda=1) to regression reward (ablation)REAL · Judge benchmarks averagePearson: 66.0 -> 67.9Improved
raise lambda from 1 to 10 (ablation)REAL · Judge benchmarks averagePearson: 67.9 -> 68.1No clear effect
restrict training to partially correct prompts (0<acc<1) rather than all prompts (ablation)REAL (K=8) · Judge benchmarks averagePearson: 65.5 -> 67.2Improved
add prediction weight beta=0.01 versus beta=0 (ablation)REAL init from SFT · Judge benchmarks averagePearson: 67.3 -> 67.9Improved
raise prediction weight beta from 0.01 to 1 (ablation)REAL init from SFT · Judge benchmarks averagePearson: 67.9 -> 66.7Hurt

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