Tight Regret Bounds for Bayesian Optimization in One Dimension

Fuente: arXiv
Saved in:
Bibliographic Details
Main Author: Scarlett, Jonathan
Format: Preprint
Published: 2018
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913823234981888
author Scarlett, Jonathan
author_facet Scarlett, Jonathan
contents We consider the problem of Bayesian optimization (BO) in one dimension, under a Gaussian process prior and Gaussian sampling noise. We provide a theoretical analysis showing that, under fairly mild technical assumptions on the kernel, the best possible cumulative regret up to time $T$ behaves as $Ω(\sqrt{T})$ and $O(\sqrt{T\log T})$. This gives a tight characterization up to a $\sqrt{\log T}$ factor, and includes the first non-trivial lower bound for noisy BO. Our assumptions are satisfied, for example, by the squared exponential and Matérn-$ν$ kernels, with the latter requiring $ν> 2$. Our results certify the near-optimality of existing bounds (Srinivas {\em et al.}, 2009) for the SE kernel, while proving them to be strictly suboptimal for the Matérn kernel with $ν> 2$.
format Preprint
id arxiv_https___arxiv_org_abs_1805_11792
institution arXiv
publishDate 2018
record_format arxiv
spellingShingle Tight Regret Bounds for Bayesian Optimization in One Dimension
Scarlett, Jonathan
Machine Learning
Information Theory
Optimization and Control
We consider the problem of Bayesian optimization (BO) in one dimension, under a Gaussian process prior and Gaussian sampling noise. We provide a theoretical analysis showing that, under fairly mild technical assumptions on the kernel, the best possible cumulative regret up to time $T$ behaves as $Ω(\sqrt{T})$ and $O(\sqrt{T\log T})$. This gives a tight characterization up to a $\sqrt{\log T}$ factor, and includes the first non-trivial lower bound for noisy BO. Our assumptions are satisfied, for example, by the squared exponential and Matérn-$ν$ kernels, with the latter requiring $ν> 2$. Our results certify the near-optimality of existing bounds (Srinivas {\em et al.}, 2009) for the SE kernel, while proving them to be strictly suboptimal for the Matérn kernel with $ν> 2$.
title Tight Regret Bounds for Bayesian Optimization in One Dimension
topic Machine Learning
Information Theory
Optimization and Control
url https://arxiv.org/abs/1805.11792