R.Riganti: Auxiliary Physics-Informed Machine Learning for Radiative Transfer and Phonon Boltzmann Transport Problems

This talk highlights recent advances in physics-informed machine learning algorithms for forward and inverse solutions of Boltzmann-type integro-differential equations. I will introduce novel, mesh-free auxiliary neural network architectures that eliminate discretization errors typically caused by quadrature evaluations of integral terms. By mapping integro-differential transport problems into equivalent differential ones, this approach significantly improves accuracy and efficiency. I will demonstrate the flexibility of the auxiliary physics-informed neural network approach through applications to both radiative transfer and phonon Boltzmann transport problems.