Stochastic Bayesian Optimization with Unknown Continuous Context Distribution via Kernel Density Estimation

Authors: Xiaobin Huang, Lei Song, Ke Xue, Chao Qian

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

Reproducibility Variable Result LLM Response
Research Type Experimental Furthermore, we conduct numerical experiments to empirically demonstrate the effectiveness of our algorithms. and In order to empirically evaluate the effectiveness of SBO-KDE and DRBO-KDE, we conduct numerical experiments on synthetic functions and two real-world problems
Researcher Affiliation Academia National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing 210023, China School of Artificial Intelligence, Nanjing University, Nanjing 210023, China {huangxb, songl, xuek, qianc}@lamda.nju.edu.cn
Pseudocode Yes Algorithm 1: SBO-KDE and Algorithm 2: DRBO-KDE
Open Source Code Yes The code is available at https://github.com/lamda-bbo/sbokde.
Open Datasets Yes We conduct numerical experiments on four commonly used synthetic test functions (Surjanovic and Bingham 2013), in which we follow the approach of setting some dimensions as context variable from (Williams 2000; Cakmak et al. 2020). and Newsvendor problem is a classic inventory management problem in stochastic optimization and We use the default setting of (Eckman et al. 2021), where s0 = 5, s1 = 9, and w = 1. and The objective function is the posterior mean of a GP trained on 3, 000 samples, which are generated by (Cakmak et al. 2020) from the CVXPortfolio problem (Boyd et al. 2017).
Dataset Splits No The paper uses an initial data set (n0 points) and then iteratively adds evaluated points, but does not specify a distinct validation set or explicit train/test/validation splits with percentages or sample counts for reproduction.
Hardware Specification No The paper does not provide any specific hardware details (e.g., GPU/CPU models, memory, or cloud instance types) used for running its experiments.
Software Dependencies No The paper mentions software like 'Bo Torch' (Balandat et al. 2020) and techniques like 'L-BFGS (Nocedal 1980)' but does not provide specific version numbers for any software dependencies or libraries needed for replication.
Experiment Setup Yes We choose the Gaussian kernel K(x) = (2π) Dc/2e x 2 2 for KDE. The bandwidth h(i) t = (4/(Dc + 2))1/(4+Dc)ˆσ(i) t t 1/(4+Dc) based on the rule of thumb (Silverman 1986), where ˆσ(i) t is the standard deviation of the ith dimension of observed context. and The radius of the distribution set is set as δt = t 2/(4+Dc).