"Computing power seems to be reaching a dead end - the end of Moore's Law. The new speed-up in Big Data and machine-learning technologies will have to come from novel mathematical and algorithmic creativity," said Anshumali Shrivastava, assistant professor of computer science at Rice University. He is passionate about solving the bottlenecks he sees as a result of the speed-up in big data and ML technologies combined with the decreasing advances in compute power.
With the aid of his 2017 CAREER Award from the National Science Foundation (NSF), Shrivastava hopes to come up with clever algorithmic strategies for undoing computational bottlenecks and enabling faster, more scalable computations. For his proposal, "Hashing and Sketching Algorithms for Resource-Frugal Machine Learning," NSF has awarded Shrivastava a five-year, $499,087 grant.
"My research leverages the existing algorithmic advances for pushing machine learning to the extreme scale. I design 'hashing and sketching algorithms,' a class of randomized algorithms that can process humongous datasets in seconds," said Shrivastava, who has joint appointments at Rice in electrical and computer engineering, and in statistics.
"Most of the machine-learning algorithms still in use were developed from the 1960s through the 1980s," he said. "They were not designed with computational complexity in mind. They focused on finding the 'right' measurements from the data. Most of the measurements are now quite expensive to compute. With Big Data, we're realizing that standard techniques fail to address new constraints of computations, energy, memory and other resources."
Shrivastava earned an integrated M.S. and B.S. in mathematics and computing from the Indian Institute of Technology, Kharagpur, in 2008, and a Ph.D. in CS from Cornell University in 2015. That year he joined the Rice faculty.
"Can we redesign current machine-learning processes that rely only on operations that are efficient and do not affect the outputs significantly? That is the central question in my research," he said.
CAREER awards support the research and educational development of young scholars likely to become leaders in their fields. The five-year grants, which are among the most competitive awarded by the NSF, are given to some 400 scholars each year across all disciplines.