Explore the groundbreaking MuZero algorithm in this 19-minute video lecture. Delve into how MuZero combines tree-based search with a learned model to achieve superhuman performance in complex, visually rich domains without prior knowledge of their underlying dynamics. Learn about its innovative approach to predicting future observations' latent representations, focusing only on task-relevant information. Discover how MuZero outperforms previous methods in 57 Atari games and matches AlphaZero's superhuman performance in Go, chess, and shogi without game rule knowledge. Gain insights into the algorithm's potential for advancing planning-based reinforcement learning in new domains where accurate environment models are unavailable.
MuZero - Mastering Atari, Go, Chess, and Shogi by Planning with a Learned Model
Overview
Syllabus
MuZero: Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model
Taught by
Yannic Kilcher
Related Courses
-
MuZero - Mastering Atari, Go, Chess, and Shogi by Planning with a Learned Model - RL Paper Explained
-
EfficientZero - Mastering Atari Games with Limited Data
-
Go-Explore: A New Approach for Hard-Exploration Problems
-
Mastering Atari with Discrete World Models - Machine Learning Research Paper Explained
-
Four Famous Deep Reinforcement Learning Projects - From Atari to Dactyl
-
Assessing Game Balance with AlphaZero - Exploring Alternative Rule Sets in Chess
