Reproduce “Navigating in High-Dimensional Search Space: A Hierarchical Bayesian Optimization Approach” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wenxuan Li, Taiyi Wang, Eiko Yoneki
HiBO (Hierarchical Bayesian Optimization) is introduced, a hierarchical algorithm that enhances high-dimensional black-box optimization by integrating global search space partitioning with local Bayesian Optimization acquisition strategies. It achieves superior performance on synthetic benchmarks and improves PostgreSQL throughput by approximately 28% over default settings in a real-world system tuning task.
Key insights
- Guiding the local optimizer's acquisition strategy with global search space partitioning information is more effective for high-dimensional black-box optimization than rigidly confining the search to a single region.
- An adaptive control mechanism for the search tree's depth, adjusting based on consecutive successes or failures, is crucial for maintaining an efficient balance between exploration and exploitation while managing computational costs.
- Combining data-driven space partitioning with UCT for evaluating partition potential allows for dynamic identification of promising regions, enabling the local optimizer to focus on areas with higher expected utility.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.