The Defense Advanced Research Projects Agency (DARPA) awarded funding to Soheil Kolouri, assistant professor of computer science, to seek ways to improve statistical modeling of machine learning systems’ outputs.
The $875,000 grant, part of DARPA’s Artificial Intelligence Exploration (AIE) Opportunity on Enabling Confidence (EC), focuses on scalable methods to generate accurate statistical models for the outputs of ML systems. Kolouri’s team, which includes Professor Hamed Pirsiavash at the University of California, Davis, submitted a proposal that allows for computationally efficient and rapid propagation of input uncertainties into the output of ML systems.
Kolouri, who directs the Machine Intelligence and Neural Technologies (MINT) Lab at Vanderbilt, said today’s ML systems fall short of having this information in their output, making them incompatible for integration with classical statistics-based estimators, which require receiving uncertainty as an input.
“DARPA EC will be a wonderful addition to our lab’s research portfolio in ML and will greatly complement our existing efforts on continual and robust ML,” he said. “Our goal is to advance fundamental research and apply our cutting-edge tools to critical applications, such as surgical operations.”
The Computer Science Department at Vanderbilt has demonstrated a strong presence in the field of ML in recent years. The Machine Learning Lunch Seminar series, co-hosted by Kolouri and Professor Tyler Derr, has drawn interest from the Vanderbilt community, the public sector, ML enthusiasts, and industrial partners.
Since the series started this spring, there have been about 20 seminars with speakers from leading national and international institutions on machine learning, and companies like Google DeepMind and Pinterest.
Contact: Lucas Johnson, 615-343-0137
lucas.l.johnson@vanderbilt.edu