$\varepsilon$-Good Action Identification in Fixed-Budget Monte Carlo Tree Search

TL;DR

This paper introduces an $\varepsilon$-agnostic algorithm for fixed-budget max-min action identification in trees, providing instance-dependent error bounds and new theoretical guarantees for Monte Carlo Tree Search.

Contribution

It presents the first provable fixed-budget algorithm with guarantees for max-min action identification, independent of $\varepsilon$, with bounds based on instance-specific gaps.

Findings

Misidentification probability decays exponentially with sample size.

The algorithm recovers known guarantees for special cases like best-arm identification.

Provides new $\varepsilon$-good guarantees for Successive Rejects.

Abstract

We study the fixed-budget max-min action identification problem in depth-2 max-min trees, an important special case of Monte Carlo Tree Search. A learner sequentially allocates $T$ samples to leaves and then recommends a subtree whose minimum leaf value is largest. Motivated by approximate planning, we focus on $\varepsilon$-good subtree identification, where any subtree whose min value is within $\varepsilon$ of the optimal maximin value is acceptable. Our main contribution is an $\varepsilon$-agnostic algorithm: it does not require $\varepsilon$ as input, but achieves instance-dependent error bounds for every meaningful $\varepsilon$. We show that the misidentification probability decays as $\exp(-\widetilde{\Theta}(T/H_2(\varepsilon)))$, where $H_2(\varepsilon)$ captures both cross-subtree and within-subtree gaps. When each subtree has a single leaf, the problem reduces to standard fixed-budget best-arm identification, and our analysis recovers, up to accelerating factors, known $\varepsilon$-good guarantees for halving-style methods while giving a new $\varepsilon$-good guarantee for Successive Rejects. On the lower-bound side, we provide complementary positive and negative results showing that max-min identification has a different hardness structure from standard $K$-armed bandits. To our knowledge, this is the first provable fixed-budget algorithmic guarantee for max-min action identification.