Gossip-based Actor-Learner Architectures for Deep Reinforcement Learning | AISC
17Speaker(s): Mido Assran
Host: Susan Shu Chang
Find the recording, slides, and more info at https://ai.science/e/gossip-based-actor-learner-architectures-for-deep-reinforcement-learning--QSSRnnwOaTVF2ByuEnPk
Motivation / Abstract
Multi-simulator training has contributed to the recent success of Deep Reinforcement Learning by stabilizing learning and allowing for higher training throughputs. We propose Gossip-based Actor-Learner Architectures (GALA) where several actor-learners (such as A2C agents) are organized in a peer-to-peer communication topology, and exchange information through asynchronous gossip in order to take advantage of a large number of distributed simulators. We prove that GALA agents remain within an epsilon-ball of one-another during training when using loosely coupled asynchronous communication. By reducing the amount of synchronization between agents, GALA is more computationally efficient and scalable compared to A2C, its fully-synchronous counterpart. GALA also outperforms A2C, being more robust and sample efficient. We show that we can run several loosely coupled GALA agents in parallel on a single GPU and achieve significantly higher hardware utilization and frame-rates than vanilla A2C at comparable power draws.
------
#AISC hosts 3-5 live sessions like this on various AI research, engineering, and product topics every week! Visit https://ai.science for more details