Speaker
Description
We present a convolutional neural network (CNN) surrogate model designed to reproduce the time-dependent synchrotron spectra of GRB prompt emission. The training set consists of physically motivated simulations in which a single electron-injection episode and a decaying magnetic field generate the evolving spectra of FRED-shaped pulses. The CNN maps the physical parameters to the full energy-and time-dependent flux, enabling rapid reconstruction of spectral evolution. Compared with our earlier CNN developed for blazar spectral energy distributions, the GRB problem requires capturing strong temporal evolution, fast changes in spectral curvature, and a tight coupling between physical parameters and time. This surrogate model will shortly be used to analyze upcoming accumulated prompt-emission data from SVOM-GRM, in addition to archival data from Fermi-GBM.
