jagomart
digital resources
picture1_Processing Pdf 179558 | 8195 On Markov Chain Gradient Descent


picture2_Processing Pdf 179558 | 8195 On Markov Chain Gradient Descent picture3_Processing Pdf 179558 | 8195 On Markov Chain Gradient Descent

 93x       Filetype PDF       File size 0.75 MB       Source: papers.neurips.cc


File: Processing Pdf 179558 | 8195 On Markov Chain Gradient Descent
OnMarkovChainGradientDescent∗ TaoSun Yuejiao Sun College of Computer Department of Mathematics National University of Defense Technology University of California, Los Angeles Changsha, Hunan 410073, China LosAngeles, CA 90095, USA nudtsuntao@163.com ...

icon picture PDF Filetype PDF | Posted on 30 Jan 2023 | 2 years ago
Partial capture of text on file.

						
									
										
									
																
													
					
The words contained in this file might help you see if this file matches what you are looking for:

...Onmarkovchaingradientdescent taosun yuejiao sun college of computer department mathematics national university defense technology california los angeles changsha hunan china losangeles ca usa nudtsuntao com sunyj math ucla edu wotaoyin abstract stochastic gradient methods are the workhorse algorithms large scale opti mization problems in machine learning signal processing and other computational sciences engineering this paper studies markov chain descent a variant where random samples taken on trajectory existing results method assume convex objec tives reversible thus have their limitations we establish newnon ergodic convergence under wider step sizes for nonconvex non nite state chains nonconvexity makes our applicable to broader problem classes hand can mix substatially faster obtain these introduce new technique that varies mixing levels reported numerical validate contributions introduction consider minimization let be statistical sample space with probability distribution omit ...
Haven't found the file you're looking for? You can try sending a request file
Comment

no comments yet
Please Login to post a comment.

no reviews yet
Please Login to review.