Stay Positive: Neural Refinement of Sample Weights

6 May 2025  ·  Benjamin Nachman, Dennis Noll ·

Monte Carlo simulations are an essential tool for data analysis in particle physics. Simulated events are typically produced alongside weights that redistribute the cross section across the phase space. Latent degrees of freedom introduce a distribution of weights at a given point in the phase space, which can include negative values. Several post-hoc reweighting methods have been developed to eliminate the negative weights. All of these methods share the common strategy of approximating the average weight as a function of phase space. We introduce an alternative approach with a potentially simpler learning task. Instead of reweighting to the average, we refine the initial weights with a scaling transformation, utilizing a phase space-dependent factor. Since this new refinement method does not need to model the full weight distribution, it can be more accurate. High-dimensional and unbinned phase space is processed using neural networks for the refinement. Using both realistic and synthetic examples, we show that the new neural refinement method is able to match or exceed the accuracy of similar weight transformations.

PDF Abstract

Categories


High Energy Physics - Phenomenology High Energy Physics - Experiment Data Analysis, Statistics and Probability