Timothy D. Kim: working at the intersection of machine learning and neuroscience

Thursday, Jul 15, 2021
by Sharon Adarlo

Timothy D. Kim, 28, doctoral student

Studies:

Kim is a doctoral student at the Princeton Neuroscience Institute, where he has been since 2015. He also recently completed the graduate certificate program at the Center for Statistics and Machine Learning.

After the first year of his doctoral studies, Kim took a leave of absence for about two years to serve in the South Korean military. During his stint, he was deployed to South Sudan, where he served as a translator and assistant instructor for a human resource development program, which was part of the United Nations Mission in South Sudan. This program was part of an effort to increase employment and mitigate tensions between different ethnic groups in South Sudan.

Before coming to Princeton, Kim earned bachelor's degrees in computer science, mathematics and biology at the University of Pennsylvania in 2015. During his undergraduate studies, he worked on developing human behavioral experiments and also on developing a computational model to characterize cognitive mechanisms underlying their behavior in the tasks that they performed.

Research:

Kim is broadly interested in computational neuroscience and machine learning.

“I am currently investigating how populations of neurons give rise to the capacity for working memory and evidence accumulation during perceptual decision-making,” he said.

In a nutshell, Kim is studying the neural systems and mechanisms behind how we make decisions that require accumulating evidence over time, such as reading articles about different cars and test driving them before buying the model of car they think best suits their needs. There are many questions on how this happens, and neuroscientists are seeking answers to these questions using a mix of computational tools and lab experiments.

For his research work, Kim takes large-scale recordings of neural populations implicated in decision-making, reduces their dimensionality, and then infers the low dimensional representation using a deep learning-based latent variable model.

However, one criticism of this data-driven modeling has been that it may be difficult to interpret the low dimensional structures inferred from this neural activity, he said.

“Improving the interpretability of those representations is currently of wide interest to neuroscientists as the availability of large-scale datasets is increasing,” Kim said.

To address this problem, he used tools from numerical differential equations and dynamical systems to interpret the low-dimensional structure. These findings will be presented at this year’s International Conference on Machine Learning in the latter part of July.

“I’m planning to apply this new method to large-scale recordings taken during experiments where rats perform a perceptual decision-making task in order to investigate what neural mechanism allows them to solve the task,” he said.

In his neuroscience work, Kim found the graduate courses at CSML to be very helpful, especially since he had limited exposure to machine learning courses in his undergraduate years.

“I took Professor Peter Ramadge’s (CSML director and professor in electrical and computer engineering) class on introduction in machine learning in 2018, and it was awesome. I think it was one of the best courses I took in Princeton. The lectures he gave were well done and very clear,” he said. “I really appreciated what I learned during that class.”

Extracurricular Activities:

Kim is a member of the Korean Graduate Student Association on campus.

For Fun:

Kim enjoys reading random Wikipedia articles, watching movies and TV shows on Netflix, and cooking.