Research map / Multi-agent systems and games / Cooperative multi-agent RL
Multi-Action Self-Improvement For Neural Combinatorial Optimization
Laurin Luttmann, Lin Xie
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 multi-agent set-based policy (MACSIM) instead of single-action SLIM, sampling | SLIM · FJSP 10x5 | makespan: 98.74 -> 97.64 | Improved |
| use MACSIM instead of SLIM, sampling | SLIM · FFSP 50x4x3 | makespan: 48.13 -> 47.74 | Improved |
| use MACSIM instead of SLIM, sampling | SLIM · HCVRP 100x3 | max route length: 10.81 -> 10.67 | Improved |
| MACSIM versus OR-Tools on FJSP 20x5 with sampling | OR-Tools · FJSP 20x5 | makespan: 188.15 -> 188.10 | Improved |
| MACSIM versus OR-Tools on FJSP 15x10 with sampling | OR-Tools · FJSP 15x10 | makespan: 143.53 vs 145.95 | Hurt |
| add autoregressive agent sampling instead of fixed order (ablation) | MACSIM · FJSP 15x10 | makespan: 155.13 -> 145.95 | Improved |
| add autoregressive agent sampling instead of random order (ablation) | MACSIM · FJSP 15x10 | makespan: 150.72 -> 145.95 | Improved |
| add skip token (ablation) | MACSIM · FJSP 15x10 | makespan: 159.65 -> 145.95 | Improved |
| add skip token (ablation) | MACSIM · FJSP 15x10 | inference time (s): 3.89 -> 6.19 | Hurt |
| add skip token (ablation) | MACSIM · FJSP 10x5 | makespan: 98.69 -> 97.64 | Improved |
| use set-based cross-entropy loss instead of single-agent cross-entropy (ablation) | MACSIM · FFSP 50x12 | makespan: 49.58 -> 47.74 | Improved |
| use set-based cross-entropy instead of L_PL (ablation) | MACSIM · FJSP 10x5 | makespan: 97.99 -> 97.64 | Improved |
Closest papers by meaning
- GOAL: A Generalist Combinatorial Optimization Agent Learner (ICLR 2025 · similarity 0.88)
- Meta$^n$: Recursive Self-Improvement through Emergent Depth (NeurIPS 2026 · similarity 0.87)
- Probing Neural Combinatorial Optimization Models (NeurIPS 2025 · similarity 0.87)
- CO-Bench: Benchmarking Language Model Agents in Algorithm Search for Combinatorial Optimization (AAAI 2026 · similarity 0.87)
- Symmetric Replay Training: Enhancing Sample Efficiency in Deep Reinforcement Learning for Combinatorial Optimization (ICML 2024 · similarity 0.86)
- PARCO: Parallel AutoRegressive Models for Multi-Agent Combinatorial Optimization (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.
