Orateur
Georgios Vasilopoulos
(National and Kapodistrian University of Athens)
Description
Leptohadronic modeling of jet emission is central to interpreting multimessenger observations, but time-dependent kinetic codes such as AM3 and LeHaMoC are too slow for Bayesian fitting over high-dimensional parameter spaces. We present GRU-based neural network emulators trained on large AM3 and LeHaMoC SED libraries. They reproduce the full broadband spectra with high accuracy at millisecond evaluation times. For blazars, this enables full posterior sampling of leptonic and leptohadronic models against multiwavelength data. We also outline the next step like extending the framework to GRBs.