Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Policy Search, Retrieval, and Composition via Task Similarity in Collaborative Agentic Systems
Saptarshi Nath, Christos Peridis, Eseoghene Benjamin, Xinran Liu, Soheil Kolouri, Peter Kinnell, Zexin Li, Cong Liu, Shirin Dora, Andrea Soltoggio
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 |
|---|---|---|---|
| add communication and sharing among agents (ablation) | MOSAIC · CT-graph ISL | maximum total return: 9.6 -> 26.0 | Improved |
| add communication and sharing among agents (ablation) | MOSAIC · MiniHack MultiRoom | maximum total return: 3.96 -> 9.04 | Improved |
| add communication and sharing among agents (ablation) | MOSAIC · MiniGrid Crossing | final return: 10.78 -> 11.67 | No clear effect |
| add cosine-similarity selection (Criterion 1) (ablation) | MOSAIC · CT-graph ISL | maximum total return: 20.1 -> 26.0 | Improved |
| add performance-based selection (Criterion 2) (ablation) | MOSAIC · CT-graph ISL | maximum total return: 25.6 -> 26.0 | No clear effect |
| add performance-based selection (Criterion 2) (ablation) | MOSAIC · CT-graph ISL | iterations to 50% of max: 71 -> 37 | Improved |
| add cosine-similarity selection (Criterion 1) (ablation) | MOSAIC · CT-graph ISL | iterations to 50% of max: 76 -> 37 | Improved |
| add reward-guided initialization (RGI) (ablation) | MOSAIC · CT-graph ISL | learning stability | Improved |
| communicate every iteration instead of every 10 (ablation) | MOSAIC · CT-graph ISL | return | Hurt |
| communicate every 25 or 40 iterations instead of 10 (ablation) | MOSAIC · CT-graph ISL | return | Hurt |
| vary embedding samples N and reference size M (ablation) | MOSAIC · CT-graph ISL | performance | No clear effect |
| use MOSAIC over MOORE | MOORE · MiniGrid Crossing | final return: 8.83 -> 11.67 | Improved |
Closest papers by meaning
- Learning to Share: Selective Memory for Efficient Parallel Agentic Systems (ICML 2026 · similarity 0.89)
- Adaptive Collaboration with Humans: Metacognitive Policy Optimization for Multi-Agent LLMs with Continual Learning (ICLR 2026 · similarity 0.88)
- Beyond Single Stationary Policies: Meta-Task Players as Naturally Superior Collaborators (NeurIPS 2024 · similarity 0.87)
- Peer Learning: Learning Complex Policies in Groups from Scratch via Action Recommendations (AAAI 2024 · similarity 0.87)
- Multi-agent cooperation through learning-aware policy gradients (ICLR 2025 · similarity 0.87)
- HyperMARL: Adaptive Hypernetworks for Multi-Agent RL (NeurIPS 2025 · 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.
