Ezequiel Alvarez: Leveraging Biased Simulations with Bayesian inference

Tuesday 2 Jun 2026, 11:00 → 12:00 Europe/Ljubljana

F1 tea room

Scientific measurements often yield data that are mixtures of signal and one or more background components. When class templates (e.g., simulation-based priors) are biased or imprecise, they are typically treated as nuisance systematics. We take a different route: by embedding imprecise priors in a hierarchical Bayesian mixture model and combining them with observed mixed data via Bayesian machine learning, we adapt class shapes toward the true distributions and obtain an unbiased posterior estimate of the signal fraction with calibrated uncertainties. In practice this lets us exploit simulation even when it is imperfect—improving template fidelity during inference. We demonstrate the approach on a collider-physics example (di-Higgs → 4b at the LHC),  and show how the same framework can enhance and unbias the ABCD background-estimation method.  The framework, Template-Adapted Mixture Model (TAMM, arXiv:2604.022219), is straightforward to implement, scales to high-dimensional and complex distributions, and is broadly applicable across HEP, astrophysics, cosmology, and any domain with mixed data and imperfect priors.