Title: Structure-preserving artificial neural networks for entropy-based moment closures
Abstract: We present how artificial neural networks can be used to substitute an otherwise costly moment closure. Specifically, we study entropy closures which have many desirable properties (among them guaranteed hyperbolicity, positivity and entropy decay). We will need to construct very specific artificial neural networks that are guaranteed to preserve these advantages. Furthermore, we address the question of data sampling. Several numerical results are shown.
(Host: Michael Herty)