Overview
Explore the groundbreaking advancements in artificial intelligence for chess and shogi in this 25-minute lecture by Kira Selby. Delve into the evolution of AI in board games, from Deep Blue to AlphaGo and AlphaZero. Examine the traditional approaches of minimax search, handcrafted features, and opening books, then contrast them with AlphaZero's revolutionary deep neural network architecture. Understand the training process, policy improvement, and the unique features that set AlphaZero apart from its predecessors. Analyze the impressive statistics and results, including training times and performance against top human players like Hikaru Nakamura. Gain insights into the future of AI in complex strategy games and its potential applications beyond the board.
Syllabus
Intro
AlphaGo
AlphaZero
Chess
Shogi
Deep Blue
Search Space
Minimax Search
handcrafted features
point valuations
opening books endgame tables
AlphaZero approach
Training of policy improvement
Deep Neural Network
Architecture
Inputs and outputs
Features
Research
Differences from AlphaGo
Statistics
Results
Training Times
Conclusion
Chess Agents
Hikaru Nakamura
Taught by
Pascal Poupart