Research map / Reinforcement learning / Offline RL
FOCAL: Efficient Fully-Offline Meta-Reinforcement Learning via Distance Metric Learning and Behavior Regularization
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 |
|---|---|---|---|
| Use inverse-power DML rather than positive-power distance losses (ablation) | FOCAL · Half-Cheetah-Vel | ESR: Square 0.506; Linear 0.819; Inverse 0.840; Inverse-square 0.861 | Helped |
| Use inverse-square DML rather than inverse DML (ablation) | FOCAL · Half-Cheetah-Vel | ESR: 0.840 -> 0.861 | Helped |
| Use inverse-square DML rather than square DML (ablation) | FOCAL · Half-Cheetah-Vel | ESR: 0.506 -> 0.861 | Helped |
| Use inverse-power DML rather than positive-power distance losses (ablation) | FOCAL · Half-Cheetah-Vel | RMS: Inverse-square 1.282; Inverse 1.217; Linear 1.385; Square 1.415 | No clear effect |
| Use a deterministic rather than probabilistic context encoder (ablation) | FOCAL · Meta-RL tasks with different reward or transition dynamics | testing return | Helped |
| Detach context-encoder training from Bellman gradients (ablation) | FOCAL · Walker-2D-Params | testing return | Helped |
| Train on expert data and test on medium data rather than matching expert data | FOCAL · Sparse-Point-Robot | average testing return: 8.16 -> 7.12 | Hurt |
| Train on expert data and test on random data rather than matching expert data | FOCAL · Sparse-Point-Robot | average testing return: 8.16 -> 4.43 | Hurt |
| Train on medium data and test on expert data rather than matching medium data | FOCAL · Sparse-Point-Robot | average testing return: 8.44 -> 8.25 | Hurt |
| Train on medium data and test on random data rather than matching medium data | FOCAL · Sparse-Point-Robot | average testing return: 8.44 -> 6.76 | Hurt |
| Use behavior regularization in offline meta-RL | FOCAL · Walker-2D-Params | Q-function divergence: >10^{8} -> approximately 200 | Helped |
Closest papers by meaning
- Towards an Information Theoretic Framework of Context-Based Offline Meta-Reinforcement Learning (NeurIPS 2024 · similarity 0.93)
- Scrutinize What We Ignore: Reining In Task Representation Shift Of Context-Based Offline Meta Reinforcement Learning (ICLR 2025 · similarity 0.92)
- Pessimistic Latent Task-aware Optimization for Robust Offline Meta-Reinforcement Learning (NeurIPS 2026 · similarity 0.91)
- Generalizable Task Representation Learning for Offline Meta-Reinforcement Learning with Data Limitations (AAAI 2024 · similarity 0.91)
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.91)
- Entropy Regularized Task Representation Learning for Offline Meta-Reinforcement Learning (AAAI 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.