»ã±¨±êÌâ (Title)£ºRecent Advances in Quasi-Newton Methods £¨ÄâÅ£¶Ù·¨µÄ×îнøÕ¹£©

»ã±¨ÈË (Speaker)£ºÂÞÂç ¸±×êÑÐÔ±£¨¸´µ©´óѧ´óÊý¾ÝѧԺ£©

»ã±¨¹¦·ò (Time)£º2023Äê11ÔÂ7ÈÕ (Öܶþ) 16:20

»ã±¨µØÖ· (Place)£ºÐ£±¾²¿GJ303

Ô¼ÇëÈË(Inviter)£ºÐì×Ë ½ÌÊÚ

Ö÷°ì²¿ÃÅ£ºÀíѧԺÊýѧϵ

»ã±¨ÌáÒª£ºWe introduce symmetric rank-$k$ methods for convex optimization to demonstrate that block quasi-Newton methods have provably faster convergence rates compared to ordinary quasi-Newton methods. We also present block Broyden's methods and square quasi-Newton methods for solving general nonlinear equations with improved convergence. For specific minimax problems, we design partial quasi-Newton methods that leverage the unbalanced dimensionality, which results in complexity matching the cost for convex minimizing problems.