AI Paper Atlas

Research map / Large language models / Alignment and preferences

Beyond Binary Preferences: A Principled Framework for Reward Modeling with Ordinal Feedback

Amirhossein Afsharrad, Ruida Zhou, Luca Viano, Sanjay Lall, Mohammad Ghavamzadeh

ICLR 2026 · Large language models › Alignment and preferences · approach: Reward model design and preference models · 1 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
Use NLL-Symmetric ordinal loss instead of Simple BTreward model, Llama-3.1-8B · RewardBench (HelpSteer2)number of errors: 433 -> 282Improved
Use NLL-Symmetric ordinal loss instead of Simple BTreward model, Llama-3.1-8B · RewardBench (HelpSteer2)mean error margin: 3.827 -> 0.501Improved
Train thresholds jointly instead of post-hoc calibration of Soft Label (ablation)ordinal reward model, Llama-3.1-8B · HelpSteer2 testMAE: 1.725 -> 1.060Improved
Train thresholds jointly instead of post-hoc calibration of Margin BT (ablation)ordinal reward model, Llama-3.1-8B · HelpSteer2 testAcc@0: 16.1% -> 29.7%Improved
Use symmetric thresholds (NLL-Sym) instead of asymmetric (NLL-Asym) (ablation)NLL ordinal loss, Llama-3.1-8B · RewardBench Avg (HelpSteer2)accuracy: 0.809 -> 0.843Improved
Use symmetric thresholds (NLL-Sym) instead of asymmetric (NLL-Asym) (ablation)NLL ordinal loss, Llama-3.1-8B · RM-Bench Total (HelpSteer2)accuracy: 0.616 -> 0.651Improved
Use NLL-Sym instead of Margin BTreward model, Llama-3.1-8B · RewardBench Avg (HelpSteer2)accuracy: 0.802 -> 0.843Improved
Use NLL-Sym instead of Margin BTreward model, Mistral-7B · RewardBench Avg (HelpSteer2)accuracy: 0.749 -> 0.779Improved
Use All-Threshold instead of Margin BTreward model, Mistral-7B · RM-Bench Total (HelpSteer3)accuracy: 0.620 -> 0.565Hurt
Use NLL-Sym instead of Scaled BTreward model, Llama-3.1-8B · RM-Bench Total (HelpSteer3)accuracy: 0.673 -> 0.676No clear effect
Corrupt 100% of labels with systematic shift noise (ablation)NLL-Symmetric · RewardBench Avgaccuracy: 0.843 -> 0.808-0.846No clear effect
Add L2 regularization on thresholds (ablation)ordinal losses · threshold convergencethreshold stabilityImproved

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