ClassBO: Bayesian Optimization for Heterogeneous Functions

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Bayesian Optimization (BO) frameworks typically assume the function to be optimized is stationary (homogeneous) over the domain. However, in many real-world applications, we often deal with functions that present a rate of variation across the input space. In this paper, we optimize functions where a finite set of homogeneous functions defined over partitions of the input space can represent the heterogeneity. The disconnected partitions that can be characterized by the same function are said to be in the same class, and evaluating the function at input returns the minimum distance to a boundary of the contiguous class (partition). The ClassGP modeling framework, previously developed to model for such heterogenous functions along with a novel ClassUCB acquisition function and partition sampling strategy, is used to introduce a novel tree-based optimization framework dubbed as ClassBO (Class Bayesian Optimization). We demonstrate the superior performance of ClassBO against other methods via empirical evaluations.

Original languageEnglish (US)
Title of host publicationLearning and Intelligent Optimization - 18th International Conference, LION 18, Revised Selected Papers
EditorsPaola Festa, Daniele Ferone, Tommaso Pastore, Ornella Pisacane
PublisherSpringer Science and Business Media Deutschland GmbH
Pages249-253
Number of pages5
ISBN (Print)9783031756221
DOIs
StatePublished - 2025
Event18th International Conference on Learning and Intelligent Optimization, LION 2024 - Ischia Island, Italy
Duration: Jun 9 2024Jun 13 2024

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume14990 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference18th International Conference on Learning and Intelligent Optimization, LION 2024
Country/TerritoryItaly
CityIschia Island
Period6/9/246/13/24

Keywords

  • Bayesian Optimization
  • Black-box Optimization
  • Gaussian process
  • Heterogeneous function
  • Non-stationary function

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science

Fingerprint

Dive into the research topics of 'ClassBO: Bayesian Optimization for Heterogeneous Functions'. Together they form a unique fingerprint.

Cite this