AlphaZero

AlphaZero learns board games by playing against itself many times.

What it does

AlphaZero learns board games by playing against itself many times.

Self-Play: Practicing with yourself • Grandmaster: Top-tier player

How it works

• Start without human game examples.
• Play games against itself.
• Keep the strategies that win more often.

Uses and applications

• It can learn games like chess or Go.
• Researchers use it to study planning and strategy.
• Game designers can study strong computer play.

In short

AlphaZero improves through self-play.

Related methods

• AlphaGo: Its predecessor, which also learned from human games.
• MCTS (Tree Search): The search AlphaZero runs before each move.
• Q-Learning: Another way to learn from rewards.

Back to Reinforcement learning, agents and planning · All topics

The interactive demonstration and the Python walkthrough for this topic need JavaScript.