Policy networks in search

[Madeleine Birchfield]

It is well known that one could replace the move ordering heuristic code in an engine's alpha-beta search with a policy neural network. What other selectivity heuristics (pruning, extensions, etc) could also be replaced by the policy network?

[smatovic]

IMO you can replace all HCSH (handcrafted search heuristics) with neural networks, it just a matter of the the extra compute time needed for nn inference during search, bigger/multiple nets vs. deeper search, a tradeoff.

--
Srdja