Research map / Reinforcement learning / Visual and representation RL
Soft Conflict-Resolution Decision Transformer for Offline Multi-Task Reinforcement Learning
Shudong Wang, Xinfei Wang, Chenhao Zhang, Shanchen Pang, Haiyuan Gui, Wenhao Ji, Xiaojian Liao
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 |
|---|---|---|---|
| add Fisher-based soft mask instead of hard mask (ablation) | SoCo-DT · MT30 near-optimal | avg success rate (%): 72.00 -> 86.33 | Improved |
| add Fisher-based soft mask instead of hard mask (ablation) | SoCo-DT · MT30 sub-optimal | avg success rate (%): 61.33 -> 69.60 | Improved |
| add IQR-based adaptive sparsity instead of fixed (ablation) | SoCo-DT · MT30 near-optimal | avg success rate (%): 76.33 -> 86.33 | Improved |
| add IQR-based adaptive sparsity instead of fixed (ablation) | SoCo-DT · MT30 sub-optimal | avg success rate (%): 59.67 -> 69.60 | Improved |
| add magnitude-sensitive harmony scoring (ablation) | SoCo-DT · MT30 near-optimal | avg success rate (%): 76.00 -> 86.33 | Improved |
| add magnitude-sensitive harmony scoring (ablation) | SoCo-DT · MT30 sub-optimal | avg success rate (%): 59.33 -> 69.60 | Improved |
| replace HarmoDT-F with SoCo-DT | SoCo-DT · MT50 sub-optimal | avg success rate (%): 57.50 -> 67.19 | Improved |
| replace HarmoDT-M with SoCo-DT | SoCo-DT · MT50 near-optimal | avg success rate (%): 78.80 -> 85.88 | Improved |
| replace HarmoDT-F with SoCo-DT | SoCo-DT · MT5 near-optimal | avg success rate (%): 100.0 -> 100.0 | No clear effect |
| replace HarmoDT-F with SoCo-DT on random 5-task combos | SoCo-DT · MT5 random combos sub-optimal | avg success rate (%): 74.67 -> 79.43 | No clear effect |
| replace HarmoDT-F with SoCo-DT on random 30-task combos | SoCo-DT · MT30 random combos near-optimal | avg success rate (%): 81.30 -> 86.27 | Improved |
| restore important parameters wrongly masked by hard masks (motivation study) (ablation) | SoCo-DT · HarmoDT on MT15 | avg success rate (%): 4.45% | Improved |
Closest papers by meaning
- Mastering Massive Multi-Task Reinforcement Learning via Mixture-of-Expert Decision Transformer (ICML 2025 · similarity 0.87)
- DaCe-DT: Data-Centric Offline Multi-Task Reinforcement Learning via Adaptive Prompts and Trajectory Correction for Heterogeneous Tasks (NeurIPS 2026 · similarity 0.87)
- HarmoDT: Harmony Multi-Task Decision Transformer for Offline Reinforcement Learning (ICML 2024 · similarity 0.87)
- AMAGO-2: Breaking the Multi-Task Barrier in Meta-Reinforcement Learning with Transformers (NeurIPS 2024 · similarity 0.86)
- Prioritized Soft Q-Decomposition for Lexicographic Reinforcement Learning (ICLR 2024 · similarity 0.86)
- TOPPO: Rethinking PPO for Multi-Task Reinforcement Learning with Critic Balancing (NeurIPS 2026 · similarity 0.86)
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.
