How Slot Machines Are Advancing the State of the Art in Computer Go AI

Explore the cutting-edge advancements in computer Go AI through this 46-minute Churchill CompSci Talk by Cheng Sun. Delve into the complexities that make Go challenging for computers and discover how the Monte Carlo Tree Search algorithm addresses these difficulties. Learn about the revolutionary impact of Monte Carlo search in 2006, which significantly outperformed traditional techniques. Examine the theory of multi-armed bandits and its application in move selection, including the asymptotically-optimal UCB1 strategy. Investigate refinements to the basic algorithm, such as AMAF/RAVE and heavy playouts, and gain insights into ongoing research topics in the field. From the game's fundamentals to the latest developments in AI strategies, this talk provides a comprehensive overview of the evolving landscape of computer Go.

Intro
Example game
Recent advancements in chess
Traditional search in go
History of go engines
Monte Carlo: philosophy
Flat Monte Carlo algorithm
MAB: greedy strategy
UCB: proof of logarithmic growth of cumulative regret
Monte Carlo simulation inconsistency
Monte Carlo Tree Search
Refinements: AMAF and RAVE
Current research