What is … a higher order tensor decomposition?
This page hosts information on Max Pfeffer's talk "What is … a
higher order tensor decomposition?" at the
"What is …?" seminar. The talk will take place on Friday, April 19, 1:00pm at the BMS Loft at
Urania. This talk will help you better understand the talk by
Reinhold Schneider, which will start at 2pm.
Abstract
Low-rank matrix structures can be exploited in many ways. We seek to generalize this concept to higher order tensors by generalizing the Singular Value Decomposition (HOSVD). Several ideas have been put forward, each proving to have certain advantages and disadvantages. Different tensor decompositions will be briefly discussed and the more recent approach of TT tensors will be introduced. The alternating least squares (ALS) algorithm will be presented as one of the most basic yet reliable tools in tensor optimization.