Research map / Reinforcement learning / Visual and representation RL
Multi-Task Reinforcement Learning with Soft Modularization
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 soft modularization with a shallow network | SAC multi-task policy · MT10-Conditioned | average success rate: 67.4 -> 71.8 | Helped |
| use soft modularization with a deep network | SAC multi-task policy · MT50-Fixed | average success rate: 35.5% -> 60.0% | Helped |
| use soft modularization with a deep network | SAC multi-task policy · MT50-Conditioned | average success rate: 34.2% -> 61.0% | Helped |
| use soft modularization with a shallow network | SAC multi-task policy · MT10-Fixed | average success rate: 85.0% -> 87.0% | Helped |
| use the deep rather than shallow module layout (ablation) | soft modularization · MT10 and MT50 Meta-World tasks | average success rate: MT10-Fixed: 87.0% -> 86.7%; MT10-Conditioned: 71.8% -> 68.4%; MT50-Fixed: 59.5% -> 60.0%; MT50-Conditioned: 60.4% -> 61. | Mixed |
| increase MT-MH-SAC depth from 3 to 4 layers (ablation) | MT-MH-SAC · MT50-Fixed | average success rate: 35.5% -> 46.7% | Helped |
| increase MT-MH-SAC depth from 4 to 5 layers (ablation) | MT-MH-SAC · MT50-Fixed | average success rate: 46.7% -> 45.2% | Hurt |
| increase MT-MH-SAC depth from 5 to 6 layers (ablation) | MT-MH-SAC · MT50-Fixed | average success rate: 45.2% -> 45.0% | Hurt |
| increase width and depth to 4 layers and 600 units (ablation) | MT-MH-SAC · MT50-Fixed | average success rate: 35.5% -> 50.7% | Helped |
| increase width and depth to 5 layers and 600 units (ablation) | MT-MH-SAC · MT50-Fixed | average success rate: 35.5% -> 50.3% | Helped |
| add task-balanced training using temperature-based weights (ablation) | Ours (Shallow) · MT10-Conditioned | average success rate | Helped |
| add observation inputs to the routing network (ablation) | Ours (Shallow) · MT10-Conditioned | average success rate | Helped |
Closest papers by meaning
- Not All Tasks Are Equally Difficult: Multi-Task Deep Reinforcement Learning with Dynamic Depth Routing (AAAI 2024 · similarity 0.89)
- Learning Policy Committees for Effective Personalization in MDPs with Diverse Tasks (ICML 2025 · similarity 0.87)
- Efficient Multi-task Reinforcement Learning with Cross-Task Policy Guidance (NeurIPS 2024 · similarity 0.86)
- Hard Tasks First: Multi-Task Reinforcement Learning Through Task Scheduling (ICML 2024 · similarity 0.86)
- Prioritized Soft Q-Decomposition for Lexicographic Reinforcement Learning (ICLR 2024 · similarity 0.86)
- AMAGO-2: Breaking the Multi-Task Barrier in Meta-Reinforcement Learning with Transformers (NeurIPS 2024 · similarity 0.85)
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.