skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Parallel Nonnegative CP Decomposition of Dense Tensors

Conference ·

The CP tensor decomposition is a low-rank approximation of a tensor. We present a distributed-memory parallel algorithm and implementation of an alternating optimization method for computing a CP decomposition of dense tensors that can enforce nonnegativity of the computed low-rank factors. The principal task is to parallelize the Matricized-Tensor Times Khatri-Rao Product (MTTKRP) bottleneck subcomputation. The algorithm is computation efficient, using dimension trees to avoid redundant computation across MTTKRPs within the alternating method. Our approach is also communication efficient, using a data distribution and parallel algorithm across a multidimensional processor grid that can be tuned to minimize communication. We benchmark our software on synthetic as well as hyperspectral image and neuroscience dynamic functional connectivity data, demonstrating that our algorithm scales well to 100s of nodes (up to 4096 cores) and is faster and more general than the currently available parallel software.

Research Organization:
Oak Ridge National Laboratory (ORNL), Oak Ridge, TN (United States)
Sponsoring Organization:
USDOE
DOE Contract Number:
AC05-00OR22725
OSTI ID:
1511940
Resource Relation:
Conference: IEEE 25th International Conference on High Performance Computing (HiPC 2018) - Bangalore, , India - 12/17/2018 3:00:00 PM-12/20/2018 3:00:00 PM
Country of Publication:
United States
Language:
English

References (23)

Constrained Tensor Factorization with Accelerated AO-ADMM conference August 2017
Time-lapse ratios of cone excitations in natural scenes journal March 2016
Tensor Decomposition for Signal Processing and Machine Learning journal July 2017
SPLATT: Efficient and Parallel Sparse Tensor-Matrix Multiplication conference May 2015
Fast Local Algorithms for Large Scale Nonnegative Matrix and Tensor Factorizations journal January 2009
The Human Connectome Project: A data acquisition perspective journal October 2012
Tensor Decompositions and Applications journal August 2009
Fast Nonnegative Matrix Factorization: An Active-Set-Like Method and Comparisons journal January 2011
Algorithms for nonnegative matrix and tensor factorizations: a unified view based on block coordinate descent framework journal March 2013
Model-Driven Sparse CP Decomposition for Higher-Order Tensors conference May 2017
Efficient algorithms for ‘universally’ constrained matrix and tensor factorization conference August 2015
Using Multivariate Analysis of Scanning-Rochigram Data to Reveal Material Functionality journal July 2016
Parallel Candecomp/Parafac Decomposition of Sparse Tensors Using Dimension Trees journal January 2018
Dynamic functional connectivity and individual differences in emotions during social stress journal September 2017
Collective communication: theory, practice, and experience journal January 2007
MPI-FAUN: An MPI-Based Framework for Alternating-Updating Nonnegative Matrix Factorization journal March 2018
Learning the parts of objects by non-negative matrix factorization journal October 1999
Positive tensor factorization journal October 2001
High Performance Parallel Algorithms for the Tucker Decomposition of Sparse Tensors conference August 2016
Optimization of Collective Communication Operations in MPICH journal February 2005
PARAFAC algorithms for large-scale problems journal May 2011
Fast Alternating LS Algorithms for High Order CANDECOMP/PARAFAC Tensor Factorizations journal October 2013
Shared-memory parallelization of MTTKRP for dense tensors conference February 2018

Similar Records

Software for Sparse Tensor Decomposition on Emerging Computing Architectures
Journal Article · Thu Jun 20 00:00:00 EDT 2019 · SIAM Journal on Scientific Computing · OSTI ID:1511940

PLANC: Parallel Low-rank Approximation with Nonnegativity Constraints
Journal Article · Fri Jun 25 00:00:00 EDT 2021 · ACM Transactions on Mathematical Software · OSTI ID:1511940

True Load Balancing for Matricized Tensor Times Khatri-Rao Product
Journal Article · Fri Jan 22 00:00:00 EST 2021 · IEEE Transactions on Parallel and Distributed Systems · OSTI ID:1511940

Related Subjects