Distributed Bayesian Matrix Decomposition for Big Data Mining and Clustering

Matrix decomposition is one of the fundamental tools to discover knowledge from big data generated by modern applications. However, it still inefficient or infeasible process very using such a method in single machine. Moreover, are often distributedly collected and stored on different machines. Thus, generally bear strong heterogeneous noise. It essential useful develop distributed matrix for analytics. Such should scale up well, model noise, address communication issue system. To this end, we propose Bayesian (DBMD) mining clustering. Specifically, adopt three strategies implement computing including 1) accelerated gradient descent, 2) alternating direction multipliers (ADMM), 3) statistical inference. We investigate theoretical convergence behaviors these algorithms. heterogeneity an optimal plug-in weighted average that reduces variance estimation. Synthetic experiments validate our results, real-world show algorithms well achieves superior competing performance compared two typical methods Scalable-NMF scalable k-means++.