Simulation-Prior Independent Neural Unfolding Procedure

2025-07-20

AI Summary

In high-energy physics, unfolding parton-level distributions is challenged by detector-induced distortions and reliance on simulated prior distributions. To address these issues, we propose SPINUP—a binless, end-to-end neural unfolding method that requires no simulated prior. SPINUP models the forward detector response via a differentiable neural network and quantifies information loss using neural importance sampling combined with ensemble learning, enabling robust inversion from high-dimensional observed space to parton-level space. Crucially, SPINUP eliminates dependence on simulated priors for the first time, substantially enhancing model independence and robustness of unfolded results. Experimental validation on jet substructure, Higgs boson associated production, and single-top quark events demonstrates superior accuracy compared to conventional simulation-based unfolding methods, while exhibiting reduced sensitivity to systematic uncertainties.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Measurement, analysis, and circumvention of Web censorship

Abstract

Machine learning allows unfolding high-dimensional spaces without binning at the LHC. The new SPINUP method extracts the unfolded distribution based on a neural network encoding the forward mapping, making it independent of the prior from the simulated training data. It is made efficient through neural importance sampling, and ensembling can be used to estimate the effect of information loss in the forward process. We showcase SPINUP for unfolding detector effects on jet substructure observables and for unfolding to parton level of associated Higgs and single-top production.

Problem

• Unfolding high-dimensional spaces without binning at LHC
• Extracting unfolded distribution independent of simulation prior
• Estimating information loss effects in forward process

Innovation

• Neural network encodes forward mapping
• Uses neural importance sampling
• Ensembling estimates information loss