MCTS as policy-improvement operator
Definition
Monte Carlo Tree Search acts as an operator that takes a policy network's initial action distribution and, by searching with a value function, returns a strictly better distribution; distilling this improved distribution back into the network amortizes search into a single forward pass and shifts the test-time scaling curve up.
Cast within
No topic-region cast yet — this appears once MCTS as policy-improvement operator's compiled claims are aligned into a topic region's argument tree.
Claims mentioning MCTS as policy-improvement operator (8)
Your notes
Loading notes...