Improved Regret Bounds for Gaussian Process Upper Confidence Bound in Bayesian Optimization

TL;DR

This paper improves the theoretical regret bounds for the GP-UCB algorithm in Bayesian optimization, showing near-optimal performance guarantees for different kernels.

Contribution

It provides tighter regret bounds for GP-UCB under Matérn and squared exponential kernels, bridging the gap with previous bounds.

Findings

Achieves O(\u221a{T}) regret for Matrn kernels.

Achieves O(r{T} \u2212 r{T} r{2}) regret for squared exponential kernels.

Refines analysis of GP-UCB's concentration behavior and information gain.

Abstract

This paper addresses the Bayesian optimization problem (also referred to as the Bayesian setting of the Gaussian process bandit), where the learner seeks to minimize the regret under a function drawn from a known Gaussian process (GP). Under a Mat\'ern kernel with a certain degree of smoothness, we show that the Gaussian process upper confidence bound (GP-UCB) algorithm achieves $\tilde{O}(\sqrt{T})$ cumulative regret with high probability. Furthermore, our analysis yields $O(\sqrt{T \ln^2 T})$ regret under a squared exponential kernel. These results fill the gap between the existing regret upper bound for GP-UCB and the best-known bound provided by Scarlett (2018). The key idea in our proof is to capture the concentration behavior of the input sequence realized by GP-UCB, enabling a more refined analysis of the GP's information gain.