Institute of Information Science, Academia Sinica

Deep reinforcement learning (DRL) has made significant progress in various fields, such as gaming, robotics, natural language processing, etc. In computer science, computer games have always been an important field to verify these DRL algorithms, where the environment is controllable and accessible compared to the complicated real-world problem. Recently, DRL algorithms such as AlphaGo, AlphaZero, and MuZero can achieve super-human performance on many computer games without using any human knowledge. This talk will first introduce the basic concept of reinforcement learning. Second, we will review the history for different reinforcement learning algorithms on computer games, especially for AlphaGo. Finally, this talk will also discuss using these DRL algorithm on other applications.