»ã±¨±êÌâ (Title)£ºDerivative-free algorithms for nonlinear least squares problems (·ÇÏßÐÔ×îÓ׶þ³ËÎÊÌâµÄÎÞµ¼Êý²½Öè)

»ã±¨ÈË (Speaker)£º·¶½ðÑà ½ÌÊÚ£¨ÉϺ£½»Í¨´óѧÊýѧ¿ÆѧѧԺ£©

»ã±¨¹¦·ò (Time)£º2023Äê11ÔÂ7ÈÕ (Öܶþ) 09:00

»ã±¨µØÖ· (Place)£ºÐ£±¾²¿GJ303

Ô¼ÇëÈË(Inviter)£ºÐì×Ë ½ÌÊÚ

Ö÷°ì²¿ÃÅ£ºÀíѧԺÊýѧϵ

»ã±¨ÌáÒª£ºIn this talk we are concerned with nonlinear least squares problems for which the exact Jacobian is not available and replaced by a probabilistic or random model. Problems of this nature arise in important practical applications, such as the data assimilation in weather prediction and the estimation of the merit function in deep learning. We will present some derivative-free algorithms for such problems and show the almost sure global convergence and complexity of the algorithms.