Novel GPU-Based Approach for Matrix Factorization using Stochastic Gradient Descent

Abstract

Recommender systems are crucial tools used in most of daily-based web applications. Matrix factorization is an advanced and efficient technique for recommender systems. Recently, Stochastic Gradient Descent (SGD) method is considered to be one of the most popular techniques for matrix factorization. SGD is a sequential algorithm, which is difficult to be parallelized for largescale problems. Nowadays, researches focus on efficiently parallelizing SGD. In this research, we propose a novel GPU approach for parallelizing SGD. The proposed approach is more efficient than recent parallel approaches because it utilizes GPU, reducing non-coalesced access of global memory and achieving load balance of threads. In addition, it does not require any sorting and/or data shuffling as preprocessing phase. Although platform used for implementation of the proposed approach is old, the implementation demonstrates 12.5x speedup over state-of-the-art GPU method, BSGD.