Lili Su
Northeastern University, MA·Computer Science·Early-career
Federated learning trains one shared machine-learning model across many devices, such as phones or sensors, without collecting their raw data in one place. This protects privacy and suits devices that must respond quickly.
- Fields
Computer Science
Mathematics & Statistics
Engineering
- Topics
federated learning
distributed machine learning
machine learning theory
- Grants
- 1 known grant, through Dec 2028
What they work on
- Build theory for federated learning that uses the statistical structure of each device's data, to explain why it succeeds in practice where optimization-only analyses predict failure.
- Replace overly pessimistic bounded gradient dissimilarity assumptions with more realistic conditions, using tools from statistical learning and high-dimensional probability.
- Design new federated learning algorithms that stay accurate despite large differences in data across devices, faulty clients and outside attacks, drawing on distributed computing and optimization.
More about the research
But the devices hold very different data, some fail or misbehave, and some may be attacked. Current theory often predicts poor results that do not match how well federated learning works in practice. This project aims to close that gap with stronger mathematical foundations and more robust algorithms.
Why it matters
Federated learning lets fields such as healthcare, self-driving vehicles, language processing and manufacturing train models without sharing private raw data. Theory that matches real performance, and algorithms that tolerate uneven data, faulty devices and attacks, could make these systems more reliable and trustworthy on edge devices.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Sanjay Purushotham
University of Maryland-Baltimore County, MD·Computer Science·Early-career
Develop multi-view federated learning algorithms that train shared models across hospitals holding different kinds of patient data, without moving the data between sites.
Hulya Seferoglu
University of Illinois Chicago, IL·Engineering·Early-career
Characterize the cost-benefit trade-offs of coded computation, which adds redundancy to tolerate slow or failed workers, in practical edge computing systems.
Xiaolei Fang
North Carolina State University, NC·Engineering
Develop a federated learning framework that lets distributed manufacturing plants jointly train condition-monitoring models while keeping each plant's data local, using robust low-rank learning for data curation and supervised dimension reduction for shared features.
The school
Northeastern University: 27 labs on PI Pages, $347M a year on research, 251 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.