Trading Convergence Rate with Computational Budget in High Dimensional Bayesian Optimization

Authors: Hung Tran-The, Sunil Gupta, Santu Rana, Svetha Venkatesh2425-2432

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We perform empirical experiments to evaluate our method extensively, showing that its sample efficiency is better than the existing methods for many optimisation problems involving dimensions up to 5000.
Researcher Affiliation Academia Hung Tran-The, Sunil Gupta, Santu Rana, Svetha Venkatesh Applied Artificial Intelligence Institute,Deakin University, Geelong, Australia {hung.tranthe, sunil.gupta, santu.rana, svetha.venkatesh}@deakin.edu.au
Pseudocode Yes Algorithm 1 MS-UCB Algorithm
Open Source Code No The paper does not provide any explicit statement or link regarding the open-sourcing of the code for the methodology described.
Open Datasets Yes We use the Gisette dataset from the UCI repository (Newman and Merz 1998) with dimension D = 5000.
Dataset Splits No The paper mentions 'validation loss' in the context of neural network parameter search, implying a validation set was used, but it does not provide specific details on the size, percentage, or methodology of the train/validation/test dataset splits for reproducibility.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory, or cloud instance types) used for running its experiments.
Software Dependencies No The paper states 'We implemented our proposed MS-UCB, Line BO, Dropout UCB and SRE in Python 3 using GPy,' but it does not specify exact version numbers for Python or GPy, which is required for reproducible software dependencies.
Experiment Setup Yes Each algorithm was randomly initialized with 20 points. To maximise the acquisition function, we used LBFGS-B algorithm with 10 D random starts. For Gaussian process, we used Matern kernel and estimated the kernel hyper-parameters automatically from data. We choose d = 5 for all methods except Line BO for which d = 1 is a requirement and the GP-UCB which works directly in original D-dimensional space. We use N0 = 1, α = 0 as parameters for our method.