HySTAR: Anchored Hypergraphs for Stable Credit Assignment in Cooperative Multi-Agent Reinforcement Learning
HySTAR targets “structural target drift,” where changing agent groupings make credit assignment unstable in cooperative MARL.
The paper proposes a MAPPO-based framework that fixes a sparse overlapping hypergraph as the value-decomposition scaffold while letting representations adapt over time. It uses spatiotemporal interaction encoding and combines temporal with structural relevance to build agent-specific advantages. The authors report gains across SMAC, GRF, Traffic Junction, and MPE, including 16.7% over MAPPO on the hardest SMAC settings. ArXiv · AI/CL/LG's note
The paper proposes a MAPPO-based framework that fixes a sparse overlapping hypergraph as the value-decomposition scaffold while letting representations adapt over time. It uses spatiotemporal interaction encoding and combines temporal with structural relevance to build agent-specific advantages. The authors report gains across SMAC, GRF, Traffic Junction, and MPE, including 16.7% over MAPPO on the hardest SMAC settings. ArXiv · AI/CL/LG's note
score 4