A Florida State University researcher aims to tackle learning efficiency challenges in training machine learning models through the help of a nearly $300,000 National Science Foundation grant.

Current technology typically sends data across internet networks for analysis by a distant “cloud” server. Edge computing is a different approach that uses edge devices, which are devices at the edge of a network, to process data much closer to where it originates, thus reducing the need for extensive data transfer. Edge devices collect large amounts of data, and federated learning — a learning framework that enables the collaboration between edge devices and a central server to train a machine learning model — is ideal for gaining useful insight from this data.

However, when federated learning is deployed in edge computing systems, a significant challenge arises — edge devices typically collect different types of data or have different hardware resources, which deteriorates the performance and efficiency of federated learning.

Xiaonan Zhang, an assistant professor in the Department of Computer Science, part of the College of Arts and Sciences, is researching how incorporating improvements to federated learning in edge computing systems could potentially solve this challenge. Zhang joined the faculty at FSU in 2020, shortly after earning her doctorate in electrical and computer engineering from Clemson University, in South Carolina.

Discovering more efficient ways to implement system-wide federated intelligence within edge computing systems will benefit industries that depend on diverse data inputs and rapid data processing such as advanced manufacturing and intelligent transportation.

In this Q-and-A, Zhang discusses her work and what the new grant means for her and FSU.

Q: Can you break down your grant-funded research?

To tackle the challenge of implementing efficient federated learning in edge computing systems, my project proposes a hardware-software co-design framework, which enables the allocation of learning models and data to the most favorable edge devices for training without wasting or overly exploiting edge resources. For instance, my framework will not dump complex video data into the nearest edge devices; instead, it will send the data to the best edge device based on the type of data and the hardware resources needed. In doing so, we expect to achieve harmony among data, edge devices, and learning models, just like an orchestra achieves harmony when different instruments come together to create beautiful music.

Q: What are the broader applications of your research?

This project will eventually enhance system-level performance to reach harmonious federated intelligence, benefiting large-scale edge computing applications such as “smart cities,” advanced manufacturing, intelligent transportation, home automation, AR/VR, and many more crucial sectors that depend on diverse data inputs and rapid data processing. In the meantime, my research will improve model accuracy, reduce computer training delays, and minimize learning costs. The insights gained are expected to also spark new advancements in the emerging fields of AI-enabled “intelligence of things” like smart locks, next-generation computing systems, and advanced communication technologies like 5G and 6G.

Q: How will this grant positively affect your goals and greater body of work?

As my first awarded federal grant as an assistant professor, this NSF grant holds particular significance to me. Being awarded this grant not only gives me more confidence but also encourages me to continue conducting research to benefit humankind. This grant serves as a pivotal moment in my academic growth, and participating in this project will be an invaluable asset to my academic development.

I feel fortunate to participate in this collaboration project with professors at Clemson University who taught me so much while discussing ideas, organizing the project and writing the proposal.

Beyond advancing the field, the grant supports the engagement and mentorship of undergraduate and graduate students in research, equipping them with the knowledge and skills necessary for their successful careers. This aligns with my longstanding goal of encouraging undergraduate students to become familiar with research and training graduate students to be excellent researchers.

To learn more about Zhang’s work and the FSU Department of Computer Science, visit cs.fsu.edu.