Amy An (BS ’24) is on a mission to change minds in computer science.

“I wanted to be a computer scientist because now, everything is computer-related,” An said. “I also heard that there are not many women in the field, so I wanted to be part of the movement to include more women in computer science.”

During her junior year, An served as treasurer of Systers, a UT student organization dedicated to recruiting and mentoring women and nonbinary individuals interested in computer science and electrical engineering. Her senior year, she was the group’s vice president.

That gave her a lot of opportunities to talk with Catherine Schuman, an assistant professor in UT’s Min H. Kao Department of Electrical Engineering and Computer Science (EECS) and the faculty advisor for Systers. An was intrigued by Schuman’s work on neuromorphic computing and spiking neural networks—computer hardware and software built to emulate biological brains—and eagerly joined her lab.

“Amy was a phenomenal student,” Schuman said. “She caught on very quickly to neuromorphic computing.”

An graduated from UT this spring with a bachelor’s degree in computer science. This summer, she began her PhD in Schuman’s laboratory.

As she continues challenging old ideas about women in STEM, An is also working to change the ‘minds’ that computer scientists create.

Uncovering How Neural Networks Form

Developing neuromorphic systems is often expensive because they are created stochastically (randomly) one connection at a time, mimicking how the neurons and synapses in your brain link up when you learn something new. That randomness makes it hard to predict the results.

“Usually, a scientist creates hundreds or thousands of neural networks, and they just throw all of them away except for that one that makes the best output,” An explained.

When An joined the Systers leadership, Schuman and EECS Dongarra Professor Michela Taufer were hoping to visualize how each neural network developed, then compare its internal structure to the quality of its outputs.

If a certain network structure tends to produce high-quality results, scientists could use that structure as a starting point instead of beginning from scratch, saving valuable time and resources.

“I joined that project because it seemed really interesting to see how neural networks are evolving, and nobody had really investigated it,” An said. “It’s sort of like archaeology, to see how those neural networks evolve and how their structural changes affect how they perform.”

“Amy has been making strong contributions to my group already,” Schuman said. “She will be continuing that effort through her graduate research and will likely also mentor future undergraduate students on the project.”

Finding a Neural Network’s Specialty

An’s PhD research will expand beyond neuromorphic ‘archaeology’ into the advantages that a neural network’s final structure might give it.

As members of TENNLab, UT’s neuromorphic computing research group, An and Schuman have access to several data sets that model applications for neuromorphic computing. These sets challenge a neural network to learn and extrapolate from a pattern—for example, giving it 50 labeled pictures of flowers, then asking it to identify flowers in 100 new photos.

An is submitting these tasks to neural networks designed for the Quantum Materials for Energy Efficient Neuromorphic Communication (Q-MEEN-C), a research group led by the University of California San Diego.

“Dr. Schuman and I are working on using our different data sets to evaluate stochastic neural networks that other physicists have created,” An said. “It’s a huge project.”
An will evaluate how well each network performs the different tasks, investigating correlations between network structure, stochasticity, output quality, and the network’s ability to ignore corrupted data (noise).

“This research is super different from what I’ve been doing as an undergrad,” An said. “I’m really excited to try new things, expand my knowledge and experience, and continue my research with Dr. Schuman.”

An is also continuing her work with Systers, serving as the treasurer for the group again this academic year. Like Schuman, she will serve as a role model for undergraduate women and nonbinary students, encouraging them to conduct research in their own passion areas of EECS.

“Undergraduate research gave me an opportunity to learn about neural networks beyond what we learn in the classroom,” An reflected. “It fills in the gaps. In a classroom, you can’t really explore every detail of those concepts. But I could actually do that in my research, and it was awesome.”

Contact

Izzie Gall (865-974-7203, egall4@utk.edu)