Answers are grounded in this paper and cite their sources.
In this paper we review basic and emerging models and associated algorithms large-scale tensor networks, especially Tensor Train (TT) decompositions novel mathematical and graphical representations. We discus the concept tensorization (i.e., creating very high-order tensors from lower-order data) and super compression of data achieved via quantized tensor (QTT) networks. The purpose of a tensorization and quantization is to, via low-rank tensor approximations "super" compression, and, compact representation of structured data. The main objective of paper is to show how tensor networks can be used to solve a wide class of data optimization problems (that are far from tractable by classical methods) by applying tensorization and performing all operations relatively small size matrices and tensors and applying iteratively and approximative tensor contractions. Keywords: Tensor networks, tensor train (TT) decompositions, matrix product (MPS), matrix product operators (MPO), basic tensor operations,, distributed representation od data optimization problems for large-scale problems: generalized eigenvalue decomposition (GEVD),/SVD, canonical correlation analysis (CCA).
No takes yet. Share an insight, caveat, or question.
Andrzej Cichocki (2014) studied this question.