Bayesian Optimization of Composite Functions

Authors: Raul Astudillo, Peter Frazier

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical experiments show that our approach dramatically outperforms standard Bayesian optimization benchmarks, reducing simple regret by several orders of magnitude.
Researcher Affiliation Collaboration 1School of Operations Research and Information Engineering, Cornell University, Ithaca, NY, USA 2Uber, San Francisco, CA, USA.
Pseudocode Yes Algorithm 1 Computation of EI-CF
Open Source Code Yes Our code is available at Astudillo (2019). URL https://github. com/Raul Astudillo06/BOCF.
Open Datasets No The paper mentions 'GP-generated test problems' and 'standard benchmark functions' (Langermann, Rosenbrock, environmental model function). While these are known functions, the paper does not provide access information (link, citation for a specific dataset instance with author/year, or repository) for pre-existing datasets used for training.
Dataset Splits No The paper states, 'For all problems and methods, an initial stage of evaluations is performed using 2(d + 1) points chosen uniformly at random over X.' However, it does not specify explicit train/validation/test dataset splits, but rather refers to sequential function evaluations.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, or memory) used for running the experiments.
Software Dependencies No The paper mentions 'independent GP prior distributions' and 'ARD squared exponential covariance function' and 'averaged version of the acquisition function', but does not list specific software dependencies with version numbers.
Experiment Setup Yes For all problems and methods, an initial stage of evaluations is performed using 2(d + 1) points chosen uniformly at random over X. As proposed in Snoek et al. (2012), for all methods we use an averaged version of the acquisition function, obtained by first drawing 10 samples of the GP hyperparameters, computing the acquisition function conditioned on each of these hyperparameters, and then averaging the results.