Combine Simulation-Based-Inference and Traditional Profile-Likelihood Results

Develop methods for performing joint statistical fits that combine simulation-based-inference results with traditional profile-likelihood workflows.

Background

Simulation-based inference (SBI) uses machine-learning methods to obtain likelihood ratios, summary statistics, or posterior densities directly from simulated samples, rather than relying on an explicitly constructed likelihood. Although SBI analyses are entering LHC workflows, they use representations and computational infrastructure that differ from conventional histogram- or parametric-model-based analyses.

The paper identifies the combination of SBI-based results with traditional profile-likelihood fits as an unresolved infrastructural issue. Solving it would enable joint statistical procedures spanning emerging machine-learning-based analyses and established HEP inference workflows in both ROOT-based and Python-based ecosystems.