Failure-Aware Gaussian Process Optimization with Regret Bounds

Authors: Shogo Iwazaki, Shion Takeno, Tomohiko Tanabe, Mitsuru Irie

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the effectiveness of F-GP-UCB in several benchmark functions, including the simulation function motivated by material synthesis experiments.
Researcher Affiliation Collaboration Shogo Iwazaki MI-6 Ltd., Tokyo, Japan iwazaki@mi-6.co.jp Shion Takeno RIKEN AIP, Tokyo, Japan shion.takeno@riken.jp Tomohiko Tanabe MI-6 Ltd., Tokyo, Japan tanabe@mi-6.co.jp Mitsuru Irie MI-6 Ltd., Tokyo, Japan irie@mi-6.co.jp
Pseudocode Yes Algorithm 1 The F-GP-UCB algorithm ... Algorithm 4 in Appendix E.3 shows the pseudo-code of our modified strategy.
Open Source Code No The paper does not provide an explicit link to open-source code for the methodology presented in this paper. It mentions GPy (a third-party library) in its references, but no code for their specific F-GP-UCB implementation.
Open Datasets Yes Synthetic experiments with the Branin function. ... We perform a numerical experiment involving quasicrystals in the Al-Cu-Mn system. ... The feasible region is the composition values that form quasicrystals based on data [16].
Dataset Splits No The paper does not explicitly provide training/test/validation dataset splits or cross-validation details for the experiments. It describes how the GP model is updated with successful observations during the optimization process.
Hardware Specification No The paper does not specify any hardware used for running the experiments (e.g., GPU models, CPU models, memory).
Software Dependencies No The paper mentions software like GPy and NLopt, but does not provide specific version numbers for these or other libraries, which are necessary for reproducible software dependencies.
Experiment Setup Yes We set βt = 2 ln(2t) as in [24] for GP-UCB and F-GP-UCB. The other parameters in F-GP-UCB are fixed as described in Sec. 5. For example, we set w = 0.75, hσ = 0.02, q = 3, θmin = 0.0001, and θmax = 0.5. ... We also fix the noise variance hyperparameter of the GP model for f to 0.0001.