Research map / Reinforcement learning / Offline RL
SMAC: Score-Matched Actor-Critics for Robust Offline-to-Online Transfer
Nathan Samuel de Lara, Florian Shkurti
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 changed | Where | Measured | Effect |
|---|---|---|---|
| pretrain offline with SMAC instead of best baseline | SAC fine-tuning · D4RL, 4 of 6 tasks | regret: 34-58% lower regret | Improved |
| pretrain offline with SMAC instead of IQL | SAC fine-tuning · D4RL six-task average | normalized regret: 0.471 -> 0.031 | Improved |
| pretrain offline with SMAC instead of CalQL/CQL | TD3 fine-tuning · D4RL six-task average | normalized regret: 0.442 -> 0.090 | Improved |
| pretrain offline with SMAC | TD3+BC fine-tuning · kitchen and walker2d | performance during fine-tuning | Hurt |
| fine-tune SMAC with TD3+BC instead of TD3 | SMAC · D4RL, six tasks | smooth transfer: 6/6 -> 4/6 environments | Hurt |
| remove RvS conditioning from diffusion score estimator (ablation) | SMAC · D4RL (Appendix E) | transfer | Hurt |
| use Adam instead of Muon in offline phase (ablation) | SMAC · D4RL (Appendix N) | transfer | Hurt |
| add Muon to baselines (ablation) | CalQL/IQL/TD3+BC · D4RL (Appendix N) | transfer stability | No clear effect |
| give baselines extra offline compute (ablation) | CalQL/IQL · Kitchen (Appendix K) | transfer gap | No clear effect |
| increase offline batch size (ablation) | SMAC · D4RL (Appendix D) | offline performance | Improved |
| increase online batch size (ablation) | SMAC · D4RL (Appendix D) | adaptation speed | Improved |
Closest papers by meaning
- Offline-Boosted Actor-Critic: Adaptively Blending Optimal Historical Behaviors in Deep Off-Policy RL (ICML 2024 · similarity 0.91)
- Adversarially Trained Weighted Actor-Critic for Safe Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.90)
- Optimistic Critic Reconstruction and Constrained Fine-Tuning for General Offline-to-Online RL (NeurIPS 2024 · similarity 0.89)
- Offline Actor-Critic Reinforcement Learning Scales to Large Models (ICML 2024 · similarity 0.89)
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- A Perspective of Q-value Estimation on Offline-to-Online Reinforcement Learning (AAAI 2024 · similarity 0.88)
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.
