My core research interest lies in developing unsupervised machine learning techniques that have both high predictive and explanatory power. My current projects include understanding the temporal evolution of Autism Spectrum Disorder and evaluating drug efficacy among diabetics using electronic health records (EHRs). While EHRs provide an inexpensive way to look at large clinical populations, they are extremely sparse and incomplete. Specifically, I am interested in models and inference procedures that can take advantage of input and intutions from experts in a flexible and robust manner.
More generally, I am interested in a variety of machine learning problems centered around Bayesian modeling and sequential decision-making. My doctoral work focused on applying Bayesian nonparametrics to reinforcement learning problems in partially observable domains. For my first masters, I developed an adaptable dialog manager for a robotic wheelchair using a planning paradigm known as partially observable Markov decision processes. During my second masters, I worked on efficient inference techniques for scaling Bayesian non-parametrics to large, real-world problems. In particular, I developed efficient inference algorithms for a latent feature model known as the Indian Buffet Process, which has applications ranging from detecting software bugs to modeling protein interactions.