Innovation at Keck: Dr. Sarah Marzen and Prediction Theory


Juliette Des Rosiers ’26
Copy Editor Intern

Dr. Sarah Marzen has been an Assistant Professor of Physics at the W.M. Keck Science Department since 2019. She received a Bachelor of Science degree in physics from the California Institute of Technology and went on to receive a Ph.D. in physics from the University of California, Berkeley.

However, Marzen’s journey into science and success in research began well before her undergraduate career. Marzen attended Thomas Jefferson High School for Science and Technology, a magnet school in Virginia that encourages students to begin pursuing their interests in science through research. Marzen embraced this jump and began work the summer after ninth grade.

“My first summer, I didn’t know what I was doing, which I think is maybe what happens to most people when they first start out doing research,” Marzen said. She continued assisting in the same lab throughout high school, applying her knowledge of The Feynman Lectures on Physics to her work.

“I thought I wanted to go into optics,” Marzen said. “One of my research projects was about negative refractive index materials and I thought that they were the coolest thing ever. I got the impression that in order to make a difference in that field, the best way to do it was to do experiments. So I wanted to do experimental optics.” This experience would go on to influence her focus in college.

“It turns out that I was horrible at experiments,” Marzen confessed. However, after rerouting to explore theoretical approaches to science, Marzen discovered an interest in the collaboration between biology and physics. She received a research opportunity from her introduction to biology class that widened her perspective on avenues for research, such as maximum entropy models for time series and ligand-gated ion channels.

In graduate school, Marzen went down the experimental path again before settling into theoretical neuroscience at the Redwood Center for Theoretical Neuroscience at the University of California, Berkeley. Nonetheless, the road to her current specialty “was a super long and windy path.”

Collaborating on multiple projects and reading articles from many academics helped guide her to prediction theory. “I collected ideas from everybody and put them in a mishmash together,” Marzen said. “I do stuff very similar to Dr. Stephanie Palmer at The University of Chicago. It’s almost like we decided to divide and conquer. We chose different systems. She’s working on flies, and I’m working on cultured neurons in humans.”

Marzen reflected on how her field lacks communication between researchers, but this was a space she felt lucky to fill. “The social situation of the field that I’m in and the adjacent fields allowed me to publish about seven papers when I was in graduate school, which is quite unusual,” Marzen said. “In most cases, you spend a long time working on a problem and it’s a crowded field. You’re lucky if you get one to three papers out of a theoretical graduate school thesis.”

Marzen published papers as a bridge between researchers who criticized each other’s work by learning everyone’s material and filling gaps in communication. Through this, her graduate experience instilled a value for collaboration and respect for her colleagues.

“I think it’s important to maintain good relationships with everybody,” she said. “To really listen to what they’re saying and to take it into account during your work”.

After receiving her doctorate, Marzen began postdoctoral work at the Massachusetts Institute of Technology in the field of prediction. Her general research question was: are organisms trying to predict their environment?

“The idea is that if you predict your environment, you get additional reward beyond what you would get if you just responded,” Marzen said. She explained that this is still just a hypothesis with her work focusing on proving its truth.

“Some people think that this is just trivially true — I don’t think it’s trivially true,” Marzen said. In addition to this preliminary hypothesis, Marzen is looking to expand her focus. In the future, she hopes to explore whether organisms are “goal-directed agents that must predict to survive.” She hopes that work in prediction theory can benefit those in multiple fields, from engineering to economics.

“If we can understand how our brains are predicting, then we can understand how to engineer prediction algorithms that are better than the previous prediction algorithms,” Marzen said. “I’ve done projects with students where we look at new engineering algorithms for prediction. I’ve also done projects with students where we analyze human data or culture neuron data to show that they’re predicting.”

A current project she is working on demonstrates how Marzen values collaboration and learning from other people’s expertise to help advance science. She is working with an experimental researcher from the University of Melbourne who gathers human data from a learning experiment.

In the study, a participant is shown a string of zeros and ones and then asked to select zero or one depending on what they thought the next digit in the sequence would be. Once a decision was made, the researcher told the participant if they had predicted correctly.

“Over time, they get better at predicting,” Marzen said. “It becomes apparent that they become near optimal prediction, given how much memory they have.”

Marzen took this experimental data and interpreted it through prediction theory. “After an analysis that was originally done by an undergraduate at CMC, it becomes clear that they’re looking for patterns and strings of symbols in order to get better at predicting,” Marzen said. “A student at Scripps found the reaction times have a relationship to the accuracy of their prediction.”

Marzen shared that she is looking for someone to join her on this project this summer and encourages anyone interested to reach out for more information!

Image Source: Sarah Marzen

Don't Miss