AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Bradley-Terry and Multi-Objective Reward Modeling Are Complementary

Zhiwei Zhang, Hui Liu, Xiaomin Li, Zhenwei Dai, Jingying Zeng, Fali Wang, Minhua Lin, Ramraj Chandradevan, Linlin Wu, Zhen Li, Chen Luo, Zongyu Wu and 3 more

ICLR 2026 · Reinforcement learning › RL from feedback · approach: Theory of RLHF · 8 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
jointly train a multi-objective head with the Bradley-Terry head (SMORM-F)single-objective reward model (gemma-2B-it) · RewardBench (UF 400k / UltraFeedback)RewardBench avg: 68.2 -> 72.8Improved
jointly train multi-objective head with BT head (SMORM-F)single-objective RM (gemma-2B-it) · RewardBench (UF 40k / HelpSteer2)RewardBench avg: 64.2 -> 71.0Improved
jointly train multi-objective head with BT head (SMORM-F)single-objective RM (Mistral-7B) · RewardBench (UF 40k / HelpSteer2)RewardBench avg: 68.5 -> 77.8Improved
add BT head to multi-objective reward model (SMORM-L)MORM (gemma-2B-it) · RewardBench (UF 400k / UltraFeedback)RewardBench score: 50.6 -> 63.6Improved
add BT head to multi-objective reward model (SMORM-L)MORM (Mistral-7B) · RewardBench (UF 40k / HelpSteer2)RewardBench score: 66.0 -> 79.9Improved
add BT head to multi-objective reward model (SMORM-L)MORM (Mistral-7B) · RM-Bench (UF 40k / HelpSteer2)RM-Bench score: 52.0 -> 64.4Improved
add BT head to MORM on Chat-Hard subsetMORM (Mistral-7B) · RewardBench Chat-Hard (UF 400k / UltraFeedback)Chat-Hard score: 65.1 -> 61.0Hurt
add BT head to MORM (binarized HelpSteer2 aggregate)MORM (Llama-3.2-3B) · RewardBench (HelpSteer2 binarized / HelpSteer2)RewardBench score: 51.3 -> 56.8Improved
add BT head to MORM (binarized UltraFeedback aggregate)MORM (Llama-3.2-3B) · RewardBench (UltraFeedback binarized / UltraFeedback)RewardBench score: 50.0 -> 56.3Improved
use SMORM-F with D_S and D_M from different datasetsSORM (Llama-3.2-3B) · RewardBench (Skywork80K / HelpSteer2)RewardBench avg: 65.8 -> 69.6Improved
use SMORM-F vs GRM (linear) w/ sft at larger dataGRM (linear) w/ sft · RewardBench (Mistral, UF 400k)RewardBench avg: 79.5 -> 79.8Improved
use SMORM reward in PPO instead of baseline rewardbaseline reward model · PPO gold score (ID)gold rewardImproved

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