Research map / Multi-agent systems and games / Cooperative multi-agent RL
TaskForce: Cooperative Multi-agent Reinforcement Learning for Multi-task Optimization
Wonhyeok Choi, Kyumin Hwang, Jihun Park, Kyoungmin Lee, Seunghun Lee, Jaeyeul Kim, Minwoo Choi, Sunghoon Im
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 |
|---|---|---|---|
| use Gram matrix of task gradients as agent observation (ablation) | TaskForce · NYU-v2 (3 tasks) | training cost / feasibility: raw-gradient variant needs ×2.59M* cost vs ×0.95 | Improved |
| add task-specific multiple agents (ablation) | TaskForce · NYU-v2 (3 tasks) | Δm: −2.89% -> −4.26% | Improved |
| add centralized critic training (ablation) | TaskForce · NYU-v2 (3 tasks) | Δm: −4.26% -> −5.23% | Improved |
| add decentralized execution (ablation) | TaskForce · NYU-v2 (3 tasks) | Δm / training cost: Δm −5.23% -> −5.18%; cost ×3.21 -> ×1.00 | Mixed |
| add gradient-based reward rG (ablation) | TaskForce · NYU-v2 (3 tasks) | Δm: −5.18% -> −6.47% | Improved |
| use TaskForce instead of Aligned-MTL | MTAN on NYU-v2 · NYU-v2 (3 tasks) | Δm: −4.93% -> −6.47% | Improved |
| use TaskForce instead of Aligned-MTL | PSPNet on Cityscapes · Cityscapes (3 tasks) | Δm: −0.02% -> −0.65% | Improved |
| use TaskForce instead of Aligned-MTL | Cityscapes semantic segmentation · Cityscapes (3 tasks) | Semseg mIoU: 67.06 -> 66.63 | Hurt |
| use TaskForce instead of NashMTL | MPNN on QM9 · QM9 (11 tasks) | Δm: +62.0% -> +59.0% | Improved |
Closest papers by meaning
- Automata-Conditioned Cooperative Multi-Agent Reinforcement Learning (ICML 2026 · similarity 0.89)
- LDSA: Learning Dynamic Subtask Assignment in Cooperative Multi-Agent Reinforcement Learning (NeurIPS 2022 · similarity 0.89)
- AMAGO-2: Breaking the Multi-Task Barrier in Meta-Reinforcement Learning with Transformers (NeurIPS 2024 · similarity 0.88)
- Learning Generalizable Skills from Offline Multi-Task Data for Multi-Agent Cooperation (ICLR 2025 · similarity 0.87)
- Learning Joint Behaviors with Large Variations (AAAI 2025 · similarity 0.87)
- Dr. MAS: Stable Reinforcement Learning for Multi-Agent LLM Systems (NeurIPS 2026 · similarity 0.87)
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.
