»ã±¨±êÌâ (Title)£ºDeep Adaptive Sampling and its Application on Surrogates£¨Éî¶È×ÔÊÊÓ¦²ÉÑù¼°ÆäÔÚ´úÀíÄ£ÐÍÖеÄÀûÓã©

»ã±¨ÈË (Speaker)£ºµÔ¼ÑÓð ÖúÀí½ÌÊÚ£¨ÉϺ£¿Æ¼¼´óѧ£©

»ã±¨¹¦·ò (Time)£º2025Äê6ÔÂ13ÈÕ£¨ÖÜÎ壩11:00

»ã±¨µØÖ· (Place)£º±¦É½Ð£ÇøGJ303

Ô¼ÇëÈË(Inviter)£º¼ÍÀö½à

Ö÷°ì²¿ÃÅ£ºÀíѧԺÊýѧϵ

ÌáÒª£ºWe present two deep adaptive sampling methods and apply one to surrogate modeling of low-regularity parametric differential equations and illustrate that this mechanism is necessary for constructing surrogate models, to deal with the sampling problem in high dimensional parameter and physics spaces, with a relatively small sample size. Both the surrogate model and sampling model are approximated with deep neural networks. In particular, the sampling model is a normalizing flow, so that the sampling is immediate. We demonstrate the effectiveness of the proposed method with a series of numerical experiments.