AlphaStar: Mastering the Real-Time Strategy Game StarCraft II
Deep reinforcement learning approaches have been shown to perform well on domains where tasks and rewards and well-defined. However, in adversarial multi-agent environments, where the agent is required to improve its policy through self-play, the agent should not only solve the given task (i.e., learning to beat itself via self-play) but also develop diverse policies and strategies over time in order to become strong and robust when playing against unseen competitors. In this talk, I will present AlphaStar which is the first AI to defeat a top professional player in the game of Starcraft, one of the most challenging Real-Time Strategy (RTS) games. Specifically, I will show how such complex and robust strategies can emerge through a distributed multi-agent RL algorithm, where a population of agents compete with each other with slightly different internal goals.
Junhyuk Oh is a research scientist at DeepMind. He received his Ph.D. from Computer Science and Engineering at the University of Michigan in 2018, co-advised by Prof. Honglak Lee and Prof. Satinder Singh. His research focuses on deep reinforcement learning problems such as dealing with partial observability, generalization, planning, and multi-agent reinforcement learning. His work was featured at MIT Technology Review and Daily Mail.