Accelerating Look-ahead in Bayesian Optimization: Multilevel Monte Carlo is All you Need

Authors: Shangda Yang, Vitaly Zankin, Maximilian Balandat, Stefan Scherer, Kevin Thomas Carlberg, Neil Walton, Kody J. H. Law

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our findings are verified numerically and the benefits of MLMC for BO are illustrated on several benchmark examples. Code is available at https://github.com/Shangda-Yang/MLMCBO. and Section 5 Numerical Results describes experiments on synthetic functions, showing figures and tables of performance.
Researcher Affiliation Collaboration 1Department of Mathematics, University of Manchester 2Meta Platforms, Inc. 3Durham University Business School.
Pseudocode Yes Algorithm 1 Bayesian optimization and Algorithm 2 Single Design with MLMC acquisition
Open Source Code Yes Code is available at https://github.com/Shangda-Yang/MLMCBO.
Open Datasets No The paper states it uses various synthetic functions from BoTorch (Balandat et al., 2020) and lists function names like Ackley, Drop Wave, Branin, Hartmann6, Cosine8. While these are well-known benchmark functions, no explicit link, DOI, or specific citation for generated datasets or their public availability is provided.
Dataset Splits No The paper does not provide specific details on training, validation, or test dataset splits, nor does it refer to predefined standard splits for the synthetic functions used.
Hardware Specification No The paper does not provide specific details about the hardware used to run its experiments.
Software Dependencies No The paper mentions BoTorch (Balandat et al., 2020) but does not provide specific version numbers for BoTorch or any other software dependencies, libraries, or solvers used in the experiments.
Experiment Setup Yes Here, we use ε = 0.2 for Figure 3(a), 3(b), 3(c), 3(d) and 3(f), and ε = 0.15 for Figure 3(e). and The initial BO run starts with 2d observations. and The number of samples in MC is N = M = 1/ε2. The number of samples in MLMC for Nl and Ml, related to ε, is chosen according to the formula in Theorem 1.