Scholarlink

The search for new physics (NP) beyond the Standard Model (BSM) involves exploring vast, complex parameter spaces. Traditional sampling methods like Markov Chain Monte Carlo (MCMC) and Hamiltonian Monte Carlo (HMC) struggle with high-dimensional, multi-modal distributions and require costly evaluations od likelihoods involving theoretical calculations and experimental data. To overcome these challenges, we propose an Machine Learning (ML)-assisted Nested Sampling (NS) approach, integrating actively trained normalizing flows and Self-Normalizing Neural Networks (SNNs) with a naive NS. This method dynamically refines the predictions, accelerates convergence, and improves sampling efficiency. Our framework provides a scalable, computationally efficient solution for BSM posterior generation, optimizing likelihood estimation and Bayesian parameter estimation while adapting to evolving experimental constraints.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

BSM Parameter Space Exploration with ML-Assisted Nested Sampling

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Rajneil Baruah,
  • Subhadeep Mondal,
  • Sunando Kumar Patra,
  • Satyajit Roy

摘要

The search for new physics (NP) beyond the Standard Model (BSM) involves exploring vast, complex parameter spaces. Traditional sampling methods like Markov Chain Monte Carlo (MCMC) and Hamiltonian Monte Carlo (HMC) struggle with high-dimensional, multi-modal distributions and require costly evaluations od likelihoods involving theoretical calculations and experimental data. To overcome these challenges, we propose an Machine Learning (ML)-assisted Nested Sampling (NS) approach, integrating actively trained normalizing flows and Self-Normalizing Neural Networks (SNNs) with a naive NS. This method dynamically refines the predictions, accelerates convergence, and improves sampling efficiency. Our framework provides a scalable, computationally efficient solution for BSM posterior generation, optimizing likelihood estimation and Bayesian parameter estimation while adapting to evolving experimental constraints.