Reinforcement Learning: AlphaGo

How AlphaGo works, based on Reinforcement Learning.
Part 2 of RL from scratch series.
• Reinforcement Learning...
0:00 - intro
0:06 - how to play Go
0:21 - introducing alphaGo
0:46 - analyzing expert games
2:17 - training an expert policy
2:47 - value functions
4:05 - search trees
5:42 - reinforcement learning
6:17 - alphaGo's value function
7:47 - alphaZero

[@ireoluwaTH]

Thank you for these rather clear explanations!

[@onhazrat]

🎯 Key Takeaways for quick navigation:
00:41 🧠 AlphaGo, the Go-playing AI, learns from human experts by analyzing prior games and then plays millions of games against itself using reinforcement learning to improve.
02:25 🤖 A policy neural network is trained to predict good moves based on the state of the Go board.
03:41 🌐 The value function estimates the likelihood of winning from a given state, helping the AI plan ahead and make strategic moves.
06:10 🔄 AlphaGo uses reinforcement learning to refine its move policy and value estimation through self-play, simulating millions of games.
07:51 🤯 AlphaZero, a newer approach, relies solely on reinforcement learning and is even more advanced, eliminating the need for learning from human experts.
Made with HARPA AI