Expected Work Search: Combining Win Rate and Proof Size Estimation
This addresses game-solving efficiency for AI researchers, offering a novel algorithm that works with or without domain-specific knowledge, though it is incremental as it builds on existing methods like Monte Carlo Tree Search and Proof Number Search.
The paper tackles the problem of game solving by proposing Expected Work Search (EWS), which combines win rate and proof size estimation to minimize expected computation, resulting in solving the empty 5x5 Go board with positional superko rules and the empty 8x8 Hex board in under 4 minutes.
We propose Expected Work Search (EWS), a new game solving algorithm. EWS combines win rate estimation, as used in Monte Carlo Tree Search, with proof size estimation, as used in Proof Number Search. The search efficiency of EWS stems from minimizing a novel notion of Expected Work, which predicts the expected computation required to solve a position. EWS outperforms traditional solving algorithms on the games of Go and Hex. For Go, we present the first solution to the empty 5x5 board with the commonly used positional superko ruleset. For Hex, our algorithm solves the empty 8x8 board in under 4 minutes. Experiments show that EWS succeeds both with and without extensive domain-specific knowledge.