about
I am a research scientist in Niko Kriegeskorte's group within the Zuckerman Institute at Columbia University. I was previously a PhD student under the supervision of Liam Paninski at the Center for Theoretical Neuroscience.
I am primarily interested in probabilistic machine learning with research experience in MCMC methods, variational inference, Bayesian nonparametrics, and Bayesian active learning.
During my PhD, I worked on MCMC methods (improving partition function estimation during simulated tempering) and improving VAEs by exploiting some mathematical properties of the ELBO. More recently, I have worked on Bayesian nonparametric methods in classifier combination. Even more recently I have been working on making inference-time scaling in LLMs more efficient.
More generally, my recent work has focused on black-box probabilistic modeling, where we can take outputs from generic ML/AI algorithms and model properties of these methods to improve their estimates or efficiency without needing to know anything about their internals. This is particularly relevant when using hosted or otherwise proprietary resources.
