Geek.AI MAgent

Geek.AI MAgent is a research platform for many-agent reinforcement learning (MARL). Unlike previous research platforms that focus on reinforcement learning research with a single agent or only few agents, MAgent aims at supporting reinforcement learning research that scales up from hundreds to millions of agents.

MAgent is able to host up to one million agents on a single GPU server. It provides environment/agent configurations and a reward description language to enable flexible environment design. Through this platform, Geek.AI has been able to demonstrate emerged collective intelligence by learning from scratch.

MAgent supports Linux and OS X running Python 2.7 or python 3 and is compatible with rule-based algorithms and deep learning frameworks.

Geek.AI is a joint research team the UK and China, with representatives from Shanghai Jiao Tong Unviersity and University College London. They are dedicated to fundamental research topics of multi-agent reinforcement learning and its potential for real-world applications through AI coordination, learning to communicate, reinforcement learning with massive numbers of agents, self-play, learning to design multi-agent environments, personality learning for AI agents, GANs for discrete data generation, and more.