Department of Computer Engineering
CS 590 SEMINAR
Stochastic Gradient Descent for Matrix Factorization Problem in Shared Memory Systems
Computer Engineering Department
Sparse matrix completion problem is finding two dense matrices such that their product is used to estimate missing values of the original matrix. It is used primarily in recommendation systems e.g. Netflix movie recommendation, Amazon product recommendation. Stochastic gradient descent (SGD) algorithm is one of the most popular methods for matrix completion problem. Applying SGD on large datasets demands high computing power, which moves the problem to the area of parallel computing. Previous studies explored parallelization schemes for SGD. Iterative and probabilistic nature of the method make it difficult to be parallelized. Some of the studies, FPSGD, HOGWILD, DSGD, proposed methods for solving issues related to synchronization and communication overhead problems of the method. In this study, we try to contribute to previous studies by focusing on partitioning models to address both load balancing and cache locality issues in shared memory systems.
DATE: 22 October, 2018, Monday, CS590 presentations begin at @ 15:40