CDS Assistant Professor Jonathan Huggins Awarded CAREER Grant from NSF

Performing well in a research setting is one thing. But when systems perform poorly in real-world settings, they can’t be trusted. A new research program aiming to remedy this shortfall has received a boost from a large award from the National Science Foundation (NSF).

Jonathan Huggins, a Boston University (BU) assistant professor in the Faculty of Computing & Data Sciences (CDS) and Department of Mathematics & Statistics, recently earned a NSF CAREER Award for his project, “Scalable and Robust Uncertainty Quantification using Subsampling Markov Chain Monte Carlo Algorithms.”

The abstract of the project proposal notes that despite empirical successes, “a lack of machine learning methods with rigorous guarantees has resulted in systems that unpredictably perform poorly in real-world settings and therefore cannot be trusted in certain areas of scientific discovery and safety-critical applications.”

“I'm really interested in developing general-purpose methods that are easy for practitioners to use while still being statistically rigorous and computationally efficient,” said Huggins. “I'm particularly interested in methods that provide accurate uncertainty quantification. Correct uncertainty quantification is often critical for scientists and other users to gain actionable insights and make optimal decisions.”

The five-year grant is under the NSF Robust Intelligence Program.

“Large datasets offer the potential to provide deep scientific and operational insights when trying to understand how complex systems function. The success of this project promises to determine how to quickly yet rigorously process such large datasets and avoid overconfidence in conclusions, given the limitations of the data and knowledge of how such systems work,” said Azer Bestavros, Warren Distinguished Professor of Computer Science and associate provost for CDS.

An Off-the-Shelf Solution for Scientists and Data Analysts

This research develops a comprehensive framework and set of algorithms scientists and data analysts could use off the shelf to address both challenges.

Huggins will develop the work within two broad-interest areas. The first enables biologists to learn about the inner workings of hard-to-observe systems, such as the internal functioning of cells or the evolutionary history of animal species. The second lets ecologists predict how ecosystems will change over months to decades, enabling better management of ecosystems and deployment of ecological monitoring efforts.

CAREER grants often have a significant educational focus. “In my case, I'm writing an accessible textbook on the design and analysis of algorithms for data science, which will be of broad interest to students and researchers in machine learning, data science, statistics, and related fields. This book is used for my math and data science course, Stochastic Methods for Algorithms,” he said. He is also co-leading a BU team developing modern introductory applied statistics courses that include a greater focus on the computational aspects of statistical inference.

Developing Fast, Trustworthy Machine Learning

Huggins is a member of BU’s Probability and Statistics research group. His research focuses on the development of fast, trustworthy machine learning and artificial intelligence methods that balance the need for computational efficiency and the desire for statistical optimality with the inherent imperfections that come from real-world problems, large datasets, and complex models.

Before joining BU, Huggins served as a postdoctoral research fellow at Harvard University’s Department of Biostatistics. He earned his Ph.D. in computer science at the Massachusetts Institute of Technology in 2018.

- Toni Fitzgerald, CDS Contributor