Benjamin M. Cobb

Nonnegative Matrix Factorization (NMF) and Non- negative Coupled Matrix Tensor Factorization (NCMTF) are Constrained Low-Rank Approximation (CLRA) models which have found use in many applications. In particular, NMF and its variants have been shown to produce high-quality soft clustering and topic modeling results with the property that each clustering assignment relates to a corresponding topic; thereby providing insight into the nature of each item in a given cluster. However, NMF and its variants are unable to process heterogeneous data represented as one or more coupled tenors. Similarly, there do not exist tensorized methods which fully preserve the afore- mentioned desirable clustering and topic modeling properties of NMF. This paper develops a higher order analog of Joint-NMF, Joint Nonnegative Coupled Matrix Tensor Factorization (Joint- NCMTF), capable of factorizing heterogeneous tensor datasets whilst fully preserving these NMF properties. To accomplish this, we develop higher-order analogs of the entire NMF process, including crucial pre and post-processing steps. By incorporating additional dimensions of information present in datasets posed as coupled higher-order tensors, our proposed Joint-NCMTF method yields higher quality clustering and topic modeling results than methods which incorporate less information. We empirically demonstrate the effectiveness of our proposed method on multiple synthetic and two real-world topic modeling tasks.

Joint Nonnegative Matrix Factorization (JointNMF) is a hybrid method for mining information from datasets that contain both feature and connection information. We propose distributed-memory parallelizations of three algorithms for solving the JointNMF problem based on Alternating Nonnegative Least Squares, Projected Gradient Descent, and Projected Gauss-Newton. We extend well-known communication-avoiding algorithms using a single processor grid case to our coupled case on two processor grids. We demonstrate the scalability of the algorithms on up to 960 cores (40 nodes) with 60% parallel efficiency. The more sophisticated Alternating Nonnegative Least Squares (ANLS) and Gauss-Newton variants outperform the first-order gradient descent method in reducing the objective on large-scale problems. We perform a topic modelling task on a large corpus of academic papers that consists of over 37 million paper abstracts and nearly a billion citation relationships, demonstrating the utility and scalability of the methods.

In this paper, several novel methods of improving the memory locality of the Sequentially Truncated Higher Order Singular Value Decomposition (ST-HOSVD) algorithm for computing the Tucker decomposition are presented. We show how the two primary computational kernels of the ST-HOSVD can be fused together into a single kernel to significantly improve memory locality. We then extend matrix tiling techniques to tensors to further improve cache utilization. This block-based approach is then coupled with a novel in-place transpose algorithm to drastically reduce the memory requirements of the algorithm by overwriting the original tensor with the result. Our approach’s effectiveness is demonstrated by comparing the multi-threaded performance of our optimized ST-HOSVD algorithm to TuckerMPI, a state-of-the-art ST-HOSVD implementation, in compressing two combustion simulation datasets. We demonstrate up to   135x reduction in auxiliary memory consumption thereby increasing the problem size that can be computed for a given memory allocation by up to   3x, whilst maintaining comparable runtime performance.