This paper investigates the expected improvement (EI) policy for optimizing deterministic objective functions within Reproducing Kernel Hilbert Spaces (RKHS). The researchers analyze the performance of EI using Gaussian process models with Matérn and squared-exponential kernels, establishing finite-budget bounds for simple regret. The findings indicate that the EI policy achieves minimax-rate optimality for Matérn kernels and near-optimality for squared-exponential kernels over RKHS balls. AI
IMPACT Provides theoretical insights into optimization strategies relevant for machine learning model training and hyperparameter tuning.
RANK_REASON This is a research paper published on arXiv detailing theoretical findings in machine learning. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
