AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

SharedRep-RLHF: A Shared Representation Approach to RLHF with Diverse Preferences

Arpan Mukherjee, Marcello Bullo, Deniz Gündüz

AAAI 2026 · Reinforcement learning › RL from feedback · approach: Theory of RLHF · 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
replace group-specific reward models with shared representation (SharedRep-RLHF)MaxMin-RLHF · IMDb sentiment, minority 1%mean minority score: 0.59 -> 0.69Improved
replace group-specific reward models with shared representationMaxMin-RLHF · IMDb sentiment (1% to 20% minority)minority win rate: 21.38 -> 25.26 (1%)Improved
replace group-specific reward models with shared representationMaxMin-RLHF · GSM8K preference, minority 5%mean minority score: 0.14 -> 0.29Improved
replace group-specific reward models with shared representationMaxMin-RLHF · GSM8K preference, minority 5%minority win rate: 4.93 -> 25.32Improved
replace group-specific reward models with shared representationMaxMin-RLHF · GSM8K preference, minority 30%mean minority score: 0.30 -> 0.27Hurt
replace group-specific reward models with shared representationMaxMin-RLHF · GSM8K preference, minority 30%minority win rate: 31.65 -> 22.76Hurt
replace group-specific reward models with shared representationMaxMin-RLHF · GSM8K preference (all minority proportions)mean minority scoreMixed

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