Timezone: »

Scalable Evaluation of Multi-Agent Reinforcement Learning with Melting Pot
Joel Z Leibo · Edgar Duenez-Guzman · Alexander Vezhnevets · John Agapiou · Peter Sunehag · Raphael Koster · Jayd Matyas · Charles Beattie · Igor Mordatch · Thore Graepel

Tue Jul 20 05:00 AM -- 05:20 AM (PDT) @ None

Existing evaluation suites for multi-agent reinforcement learning (MARL) do not assess generalization to novel situations as their primary objective (unlike supervised learning benchmarks). Our contribution, Melting Pot, is a MARL evaluation suite that fills this gap and uses reinforcement learning to reduce the human labor required to create novel test scenarios. This works because one agent's behavior constitutes (part of) another agent's environment. To demonstrate scalability, we have created over 80 unique test scenarios covering a broad range of research topics such as social dilemmas, reciprocity, resource sharing, and task partitioning. We apply these test scenarios to standard MARL training algorithms, and demonstrate how Melting Pot reveals weaknesses not apparent from training performance alone.

Author Information

Joel Z Leibo (DeepMind)
Edgar Duenez-Guzman (DeepMind)
Sasha Vezhnevets (DeepMind)
John Agapiou (DeepMind)
Peter Sunehag
Raphael Koster (DeepMind)
Jayd Matyas (DeepMind)
Charlie Beattie (DeepMind Technologies Limited)
Igor Mordatch (Google Brain)
Thore Graepel (DeepMind)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors