A Colby Grad Gets Noticed for Her Research in Machine Learning

Alumni4 MIN READ

Natalie Maus ’21 looked at the edge humans have over machines when it comes to navigating in dark conditions

Natalie Maus ’21 poses for a portrait at the University of Pennsylvania where she’s studying for a doctorate in Computer Science.
By Abigail Curtis Photography by Gabe Souza
November 17, 2022

Neural Networks, a peer-reviewed scientific journal, recently published the senior thesis of a Colby graduate whose research compares how well machines and humans determine where they are going when navigating through their environment.

Natalie Maus ’21 measured the degree to which humans have the edge over machines, a  question that matters a lot in a world where artificial intelligence is becoming ever more important. Oliver Layton, assistant professor of computer science who served as her thesis advisor, said that as technology evolves, machines will need to improve in this regard, so it’s critical to understand their capacity. 

“I think her research makes a significant contribution,” he said. “It provides an answer to a very useful question—when it comes to navigation and perceiving where we’re going, how well do humans and machines perform relative to each other? Especially in this era of self-driving cars, how feasible is it to replace a human with a machine? This provides a really detailed comparison.” 

Maus, now in the second year of a doctoral program at the University of Pennsylvania, described herself as “definitely a science nerd” who wants most of all to do something that makes a difference. 

“I want to spend the rest of my career working on problems that I think are exciting and trying to advance research, and leave behind a world where scientists know more than they did before.”

Natalie Maus ’21

The Colorado native figured she’d major in math at Colby, but after taking an intro to computer science course and loving it she switched directions. Maus, who played ice hockey for the Colby Mules, ended up double majoring in physics and computer science.

“The reason why I went down the computer science-machine learning route is that I think that machine learning right now is able to help make progress in other areas of science,” she said. 

Toward that end, she did a summer internship at NASA’s Jet Propulsion Laboratory. Because of the pandemic, it was remote, but she “learned a ton” about the research process anyway, she said. 

The pandemic almost got in the way of her senior honors thesis, too. It was during the height of the pandemic that she approached Layton to ask if he would be her advisor. Initially, he was doubtful—he hadn’t had her as a computer science student and wasn’t planning to take on new students. But after meeting Maus, his doubts dissipated. 

“She’s highly motivated, very well organized, and very bright. It was a total pleasure working with her,” he said. “She asked the right questions and is very goal-oriented. She knew exactly what she wanted to do and always had a clear direction.”

A neural network is a method in artificial intelligence that teaches computers to process data in a way that is inspired by the human brain. Maus wanted to figure out how well this kind of network uses visual stimuli to perceive a common task like walking or driving through the world. It’s something that humans are very good at, even in dimly lit environments. 

“Give me a glimmer of light, and I’ll get to where I’m going,” Layton said. 

Through her research, which was mostly done with simulated computer models, some of which looked like a state-of-the-art video game, Maus learned that machines can’t always do that. Although she was able to get them somewhat close, humans still had the edge in dark environments. That’s important because machines such as self-driving cars need to be able to estimate their direction of travel as well as people can.  

Natalie Maus
Natalie Maus ’21 works on code having to do with protein structures at the University of Pennsylvania, where she’s studying for a doctorate in computer science.

“There’s a lot of hype, but maybe we’re putting the cart in front of the horse a little bit,” Layton said. 

Maus’s thesis was featured on Undergraduate Research Commons, a website that showcases outstanding undergraduate research. When Maus and Layton submitted it to Neural Networks, they were required to do a lot of extra simulations before it was accepted for publication. The journal is well-regarded, Layton said, and is considered to have a large impact in the field. He also recently virtually presented their work at Brown University’s Perception & Action Seminar Series. 

The thesis’s success in the greater world does not surprise him. 

“Natalie has a rare sophistication and attention to detail,” he said. “It is incredible what she was able to do in a short period of time, especially in the pandemic. It is definitely a story of endurance.” 

Maus said that it was very exciting to have her thesis published. 

“The goal should be to move forward in science as well as you can, not just write papers,” she said. “But at the same time, writing papers motivates people to share their work and take the next step.” 

As for her own next steps, she is contemplating career paths such as becoming a professor or working in a big industry research lab. For the moment, though, she is simply enjoying her doctoral program. 

“I absolutely love it,” she said. “I’m really happy that this is what I decided to do.”