Explore a groundbreaking algorithm that unifies approaches for both perfect and imperfect information games in this 54-minute video featuring special guest Martin Schmid. Dive into the Player of Games algorithm, which combines guided search, self-play learning, and game-theoretic reasoning to achieve strong performance across various game types. Learn about tree search algorithms, counterfactual value and policy networks, and the unique search procedure employed by Player of Games. Discover how this innovative approach performs in chess, Go, poker, and Scotland Yard, demonstrating its potential as a general-purpose algorithm for diverse gaming environments.
Player of Games - All the Games, One Algorithm!
Overview
Syllabus
- Introduction
- What games can Player of Games be trained on?
- Tree search algorithms AlphaZero
- What is different in imperfect information games?
- Counterfactual Value- and Policy-Networks
- The Player of Games search procedure
- How to train the network?
- Experimental Results
- Discussion & Outlook
Taught by
Yannic Kilcher
Related Courses
-
ReBeL - Combining Deep Reinforcement Learning and Search for Imperfect-Information Games
-
DeepMind's AlphaGo Zero and AlphaZero - RL Paper Explained
-
From Tic Tac Toe to AlphaGo - Playing Games with AI and Machine Learning
-
MuZero - Mastering Atari, Go, Chess, and Shogi by Planning with a Learned Model
