Body

Shrivastava, Baraniuk get NSF grant for machine learning

Rice University engineers have been awarded a National Science Foundation grant to extend the foundations of Randomized Numerical Linear Algebra.

Shrivastava,  Baraniuk get NSF grant for machine learning

Rice University engineers have been awarded a National Science Foundation grant to extend the foundations of Randomized Numerical Linear Algebra (RandNLA), a vital new tool for machine learning, statistics and data analysis.

Anshumali Shrivastava, assistant professor of computer science, Rich Baraniuk, the Victor E. Cameron Chair in Engineering and director of Open Stax along with Michael Mahoney from the University of California, Berkeley, have received a four-year, $1.2-million grant.

“We’re excited to be launching a collaboration with Mike Mahoney of U.C. Berkeley in this project, plus having ONR (Office of Naval Research) involved in funding some other aspects,” Baraniuk said.

The proposal is titled “BIGDATA: F: Collaborative Research: Theory and Practice of Randomized Algorithms for Ultra-Large-Scale Signal Processing.” The focus is blending efficient hashing algorithms with RandNLA.

Managing big data has become increasingly challenging. The ability to observe massive amounts of data coming from distributed and disparate high-resolution sensors has been instrumental in enhancing the understanding of many physical phenomena. Signal processing has been the driving force in this knowledge. In the last decade, the increase in observations has outpaced computing abilities to process, understand and organize this massive but useful data.

Baraniuk and Shrivastava hope to achieve two complementary goals. First, they want to extend the foundations of RandNLA by tailoring randomization directly towards downstream end-goals provided by the underlying problem, rather than intermediate matrix approximations goals.

Secondly, they hope to use the statistical and optimization insights obtained from downstream applications to transform and extend the foundations of RandNLA.

“There is more or less a consensus that due to prohibitive resource requirements, existing machine learning and signal processing algorithms are not going to survive for long,” Shrivastava said.

“With the end of Moore’s Law, advances in the future must come from smarter algorithms exploring, which is the primary goal of this project. There is no free lunch, which means we have to trade something to obtain resource efficient algorithms. It turns out that we can trade a negligible amount of certainty, using the idea of randomization, for tremendous, often exponential gains,” he said.

The NSF has awarded six new grants to the Rice CS department since June 1, providing $5.2 million in research funding for the next four years.

Cintia Listenbee, Communications and Marketing Specialist in Computer Science