Ben Cobb

Welcome

Hi! I am currently a PhD student at the Georgia Institute of Technology under the supervision of Haesun Park and Richard Vuduc. Previous to attending Georgia Tech, I attended Wake Forest University, where I received my Bachelor of Sciences’ in Computer Science and Mathematical Business. To get in contact with me, feel free to send me an email using the link below.

Research

My current research focuses on developing distributed Non-negative Matrix Factorization (NMF) algorithms within the Parallel Low-rank Approximations with Non-negativity Constraints (PLANC) framework for use in large-scale datasets. This work is in close collaboration with Oak Ridge National Lab (ORNL).

Previous to this, I worked under the supervision of Ümit V. Çatalyürek on the GenTen project in collaboration with Sandia National Laboratories (SNL) as part of the Department of Energy’s (DOE) larger Exascale Computing project. On this project, I had the pleasure of collaborating with Eric Phipps and Hemanth Kolla as we worked to optimize portable dense tensor kernels based upon the Kokkos programming model for use in compressing large combustion simulation datasets. As part of this, we developed a novel in-place variant of the Sequentially Truncated Singular Value Decomposition (ST-HOSVD) for computing the Tucker Decomposition. The resulting algorithm, referred to as the Fused In-place Sequentially Truncated Singular Value Decomposition (FIST-HOSVD), was shown to reduce the memory consumption of computing the Tucker Decomposition by ~135x with comparable runtime performance compared to the current state-of-the-art ST-HOSVD implementation within the TuckerMPI framework.

My undergraduate research primarily focused on applying wedge-sampling heuristics to speeding up the coarsening phase of hypergraph partitioners and applying bipartite matching algorithms to computing least-squares cosine differences of ktensors. This work was performed under the supervision of my undergraduate advisor and mentor, Grey Ballard.

Selected Publications

FIST-HOSVD: Fused In-place Sequentially Truncated Higher Order Singular Value Decomposition

Cobb, Benjamin, Kolla, Hemanth, Phipps, Eric, and Çatalyürek, Ümit V.

In Platform for Advanced Scientific Computing (PASC’22) 2022

Abs

In this paper, several novel methods of improving the memory lo- cality 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 com- putational 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 re- quirements of the algorithm by overwriting the original tensor with the result. Our approach’s effectiveness is demonstrated by compar- ing the multi-threaded performance of our optimized ST-HOSVD algorithm to TuckerMPI, a state-of-the-art ST-HOSVD implemen- tation, in compressing two combustion simulation datasets. We demonstrate up to 135x reduction in auxiliary memory consump- tion thereby increasing the problem size that can be computed for a given memory allocation by up to 3x, whilst maintaining comparable runtime performance.