# Adaptive Candidate Point Thompson Sampling for High-Dimensional Bayesian Optimization

Source: [Adaptive Candidate Point Thompson Sampling for High-Dimensional Bayesian Optimization](<https://www.sota2.com/research/paper/69dc78577d0172140ec7889a>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-14T05:00:04.234Z

Authors: Donney Fan, Geoff Pleiss

Published: 2026-04-10T02:54:31.000Z

## Abstract

In Bayesian optimization, Thompson sampling selects the evaluation point by sampling from the posterior distribution over the objective function maximizer. Because this sampling problem is intractable for Gaussian process (GP) surrogates, the posterior distribution is typically restricted to fixed discretizations (i.e., candidate points) that become exponentially sparse as dimensionality increases. While previous works aim to increase candidate point density through scalable GP approximations, our orthogonal approach increases density by adaptively reducing the search space during sampling. Specifically, we introduce Adaptive Candidate Thompson Sampling (ACTS), which generates candidate points in subspaces guided by the gradient of a surrogate model sample. ACTS is a simple drop-in replacement for existing TS methods -- including those that use trust regions or other local approximations -- producing better samples of maxima and improved optimization across synthetic and real-world benchmarks.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [High-dimensional optimization on Median Molecules 256D 2](<https://www.sota2.com/research/sota/high-dimensional-optimization-on-median-molecules-256d-2.md>): High-dimensional optimization; Median Molecules 256D 2; Objective Value: 0.14.

- [High-dimensional optimization on Median Molecules 256D 1](<https://www.sota2.com/research/sota/high-dimensional-optimization-on-median-molecules-256d-1.md>): High-dimensional optimization; Median Molecules 256D 1; Objective Value: 0.3.

- [High-dimensional optimization on Rover 60D](<https://www.sota2.com/research/sota/high-dimensional-optimization-on-rover-60d.md>): High-dimensional optimization; Rover 60D; Objective Value: 2.87.

- [High-dimensional optimization on LassoDNA 180D](<https://www.sota2.com/research/sota/high-dimensional-optimization-on-lassodna-180d.md>): High-dimensional optimization; LassoDNA 180D; Objective Value: -0.28.

- [High-dimensional optimization on MOPTA08 124D](<https://www.sota2.com/research/sota/high-dimensional-optimization-on-mopta08-124d.md>): High-dimensional optimization; MOPTA08 124D; Objective Value: -215.81.

- [High-dimensional optimization on SVM 388D](<https://www.sota2.com/research/sota/high-dimensional-optimization-on-svm-388d.md>): High-dimensional optimization; SVM 388D; Objective Value: 0.04.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.08891>)

[PDF](<https://arxiv.org/pdf/2604.08891.pdf>)
