New Hierarchical Bayesian Quadrature method improves numerical integration

Researchers have developed Hierarchical Bayesian Quadrature (HBQ), a novel method for numerical integration that addresses the limitations of standard Bayesian Quadrature when dealing with nonstationary integrands. HBQ adaptively partitions the integration domain into local stationary models, recombining estimates through a GP conditioning hierarchy to capture cross-subdomain correlations. This approach avoids Markov Chain Monte Carlo (MCMC) and adjusts its computational budget based on local integrand complexity, showing significant improvements on benchmark problems and a real-world epidemiological model. AI

IMPACT This method could enhance the accuracy and efficiency of integral estimations in probabilistic machine learning and scientific computing.

RANK_REASON The cluster contains a research paper detailing a new methodology in numerical integration. [lever_c_demoted from research: ic=1 ai=0.7]

Read on arXiv cs.LG →

• Bayesian Quadrature
• Gaussian process
• Hierarchical Bayesian Quadrature

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →