Randomized algorithms for the approximation of matrices
Wednesday, September 28, 2011 - 2:00pm - 3:00pm
I will discuss recent algorithmic developments for the classical problem of approximating a given matrix by a low-rank matrix. This is motivated by the need of faster algorithms for very large data and certain applications that want the approximating matrix to have rows living in the span of only a few rows of the original matrix, which adds a combinatorial twist to the problem. The novel algorithms are based on sampling rows randomly (but non-uniformly) and random projection, from which a low rank approximation can be computed.