Speaker
Description
Modeling the spectral energy distributions (SEDs) of blazars with physically motivated models is computationally expensive, as it requires solving coupled differential equations numerically and scanning high-dimensional parameter spaces.
In this contribution I will present our recent application of machine learning to accelerate the evaluations of blazar SEDs. Our method relies on a neural network (NN) architecture based on Gated Recurrent Units (GRUs), trained on a large sample of lepto-hadronic blazar simulations computed with the publicly available codes $AM^3$ and $LeHaMoC$. The resulting NN offers a strongly reduced run time while maintaining high accuracy in SED prediction. This computational speed enables Bayesian inference to be performed efficiently, making the method suitable for the analysis of the multi-messenger data of blazars.