Random Exploration in Bayesian Optimization: Order-Optimal Regret and Computational Efficiency
Authors: Sudeep Salgia, Sattar Vakili, Qing Zhao
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We compare the computational efficiency of REDS against algorithms with order-optimal regret performance, namely BPE (Li & Scarlett, 2022) and GP-Thre DS (Salgia et al., 2021) through an empirical study. |
| Researcher Affiliation | Collaboration | 1Department of Electrical and Computer Engineering, Carnegie Mellon University, Pittsburgh, PA, USA 2Media Tek Research, Cambridge, UK 3Department of Electrical and Computer Engineering, Cornell University, Ithaca, NY, USA. |
| Pseudocode | Yes | Algorithm 1 Random Exploration with Domain Shrinking |
| Open Source Code | No | The paper does not contain an explicit statement about releasing source code or a link to a code repository for the described methodology. |
| Open Datasets | No | The paper utilizes common benchmark functions (Branin, Hartmann-4D, Hartmann-6D) for simulation experiments, which are mathematical definitions rather than pre-existing datasets. It does not provide access information (links, DOIs, or citations to specific data files) for any publicly available or open dataset. |
| Dataset Splits | No | The paper describes a sequential optimization problem where points are chosen sequentially and observations are made. It does not refer to traditional dataset splits like training, validation, and test sets, which are typical for supervised learning. Therefore, no specific split percentages or sample counts are provided. |
| Hardware Specification | No | The paper discusses 'running time' and 'computational cost' but does not provide specific details about the hardware used to run the experiments, such as CPU or GPU models, or memory specifications. |
| Software Dependencies | No | The paper mentions using a 'Square exponential kernel' and 'Gaussian noise' for its experiments, but it does not specify any software dependencies with version numbers (e.g., Python version, specific libraries like PyTorch or TensorFlow versions). |
| Experiment Setup | Yes | For all the experiments, we use the Square exponential kernel. The length scale was set to 0.2 for Branin and 1 for Hartmann-4D and Hartmann-6D functions. We corrupted the observations with a zero mean Gaussian noise to the with a standard deviation of 0.2. The value of τ was also set to 0.2. All the algorithms were run for T = 1000 time steps... The parameter N1 in REDS and BPE was set to 50 for Branin and 100 for Hartmann-4D and Hartmann-6D functions. |