Max-value Entropy Search for Efficient Bayesian Optimization

Authors: Zi Wang, Stefanie Jegelka

ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we probe the empirical performance of MES and add-MES on a variety of tasks.
Researcher Affiliation Academia 1Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology, Massachusetts, USA. Correspondence to: Zi Wang <ziw@csail.mit.edu>, Stefanie Jegelka <stefje@csail.mit.edu>.
Pseudocode Yes Algorithm 1 Max-value Entropy Search (MES)
Open Source Code Yes Our Matlab code and test functions are available at https://github.com/ zi-w/Max-value-Entropy-Search/.
Open Datasets Yes We test regression on the Boston housing dataset and classification on the breast cancer dataset (Bache & Lichman, 2013).
Dataset Splits Yes Both of the datasets were randomly split into train/validation/test sets.
Hardware Specification Yes All the timing experiments were run exclusively on an Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz.
Software Dependencies No The paper mentions "Our Matlab code" but does not specify the Matlab version or any other software dependencies with specific version numbers.
Experiment Setup Yes The parameter for GP-UCB was set according to Theorem 2 in (Srinivas et al., 2010); the parameter for PI was set to be the observation noise σ. For the functions with unknown GP hyper-parameters, every 10 iterations, we learn the GP hyper-parameters using the same approach as was used by PES (Hern andez-Lobato et al., 2014).