[BANANA] Post-seminar talk on tensors

[Mark Hoemmen]

Greetings!

This Wednesday at noon (immediately following the LAPACK seminar) in
the Wozniak Lounge (4th floor) in Soda Hall, there will be a talk on
tensors that may be of interest.  Details are below.

mfh

Sampling in large matrices, tensors.

Ravi Kannan, Yale University

Noon, Wozniak Lounge, Wed, Oct 24.

For problems with massive input data which cannot
be stored in RAM, sampling is a standard approach.
Here we first consider matrix problems. We will
show that given a small random subset of rows of
any matrix and a small random subset of columns,
one can approximate the matrix. The sampling
probabilities have to be judiciously chosen.
These approximations are relevant for Principal
Component Analysis and other applications. We also
discuss problems where the data forms a multi-dimensional
array (tensor) and Linear Algebra is not available.
Here we develop sampling based algorithms for finding
``low-rank'' approximating tensors which help us
tackle a class of combinatorial problems known as
Constraint Satisfaction Problems.