Shashanka Ubaru (University of Minnesota)

Aug 22, 2016.

Title and Abstract

Using error correcting codes for low rank approximation of large matrices and group testing
Low rank approximation is an important tool used in many applications of signal processing and machine learning. Recently, randomized sketching algorithms were proposed to effectively construct low rank approximations and obtain approximate singular value decompositions of large matrices. In the first part of this talk, we will discuss how matrices from error correcting codes can be used to find such low rank approximations and matrix decompositions, and extend the framework to linear least squares regression problems. The theoretical results that establish the desired approximation guarantees will be discussed, including the various benefits of these code matrices over other sampling matrices in different computational environments.

Despite a large volume of research in group testing, explicit small-size group testing schemes are still difficult to construct. In the second part of this talk, we will discuss an experimental study of almost disjunct matrices constructed from low-weight codewords of binary BCH codes, and evaluate their performance in nonadaptive group testing. Estimates of the error probability of identification in these constructed schemes which provides a partial explanation of their performance will also be discussed.

Bio

Shashanka Ubaru is a Ph.D student in the Department of Computer Science and Engineering, at University of Minnesota (UMN), advised by Prof. Yousef Saad. He also works with Prof. Arya Mazumdar, who co-advised his master's thesis. His research interests are in Numerical Linear Algebra, Machine Learning and Signal Processing. In particular, he is interested in the use of computational linear algebra tools for solving problems related to machine learning, data analysis and signal processing. He is also interested in the applications of error correcting codes in machine learning, and in material informatics