Research map / Reinforcement learning / RL from feedback
Direct Preference-based Policy Optimization without Reward Modeling
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| add smoothness regularizer nu to PT+IQL | PT+IQL · D4RL pen-human | normalized return: 53.0 -> 53.7 | Helped |
| add smoothness regularizer nu to PT+IQL | PT+IQL · D4RL pen-cloned | normalized return: 42.9 -> 49.8 | Helped |
| add smoothness regularizer nu to PT+IQL | PT+IQL · D4RL kitchen-partial | normalized return: 40.2 -> 49.4 | Helped |
| add smoothness regularizer nu to PT+CQL | PT+CQL · D4RL pen-human | normalized return: 31.6 -> 18.3 | Hurt |
| add smoothness regularizer nu to PT+CQL | PT+CQL · D4RL pen-cloned | normalized return: 18.3 -> 32.7 | Helped |
| add smoothness regularizer nu to PT+CQL | PT+CQL · D4RL Adroit pen and Kitchen averages | normalized return: 25.0 -> 25.5 | Mixed |
| replace reward modeling plus RL with direct preference optimization (DPPO) instead of PT+IQL | DPPO · D4RL Adroit pen average | normalized return: 48.0 -> 75.7 | Helped |
| replace PT+IQL with DPPO | DPPO · D4RL Gym average | normalized return: 67.8 -> 78.8 | Helped |
| replace PT+CQL with DPPO | DPPO · D4RL Gym average | normalized return: 62.2 -> 78.8 | Helped |
| use DPPO instead of PPO for RLHF fine-tuning | OPT-1.3b · HH-RLHF | avg reward: 4.335 -> 4.515 | Helped |
| use DPPO instead of PPO for RLHF fine-tuning | OPT-1.3b · HH-RLHF | KL divergence: 0.0091 -> 0.0083 | Helped |
| use DPPO instead of PPO for RLHF fine-tuning | OPT-1.3b · HH-RLHF | human eval win rate: 0.667 -> 0.697 | Helped |
Closest papers by meaning
- Provable Reward-Agnostic Preference-Based Reinforcement Learning (ICLR 2024 · similarity 0.95)
- Decoding Global Preferences: Temporal and Cooperative Dependency Modeling in Multi-Agent Preference-Based Reinforcement Learning (AAAI 2024 · similarity 0.92)
- Policy-labeled Preference Learning: Is Preference Enough for RLHF? (ICML 2025 · similarity 0.92)
- RA-PbRL: Provably Efficient Risk-Aware Preference-Based Reinforcement Learning (NeurIPS 2024 · similarity 0.91)
- Provable Offline Preference-Based Reinforcement Learning (ICLR 2024 · similarity 0.91)
- Adversarial Policy Optimization for Offline Preference-based Reinforcement Learning (ICLR 2025 · similarity 0.91)
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.