MARL for Combinatorial Optimization

Decentralized training seems to improve sample complexity for