Research

To improve the efficiency of Monte Carlo integration, this study proposes a machine learning algorithm for stratified sampling that divides the domain space based on the integrand's value, analogous to Lebesgue integration. By utilizing neural networks to learn complex isocontours and preclassify the space, the framework significantly reduces variance in multidimensional integration. This methodology has proven highly effective when applied to high-energy physics, facilitating faster and more precise scattering amplitude calculations and event generation.