I build machine vision methods that biomedical practitioners can use as-is to understand their data without needing to retrain or adapt to their use-cases.

Previously, I was a postdoc at MIT CSAIL in Polina Golland’s Medical Vision Group where I worked on representation learning frameworks for automatic generalization to entirely new tasks, biomedical contexts, and datasets (the trick was writing our own data engines). I got my Ph.D. from NYU under Guido Gerig at the VIDA Center where I worked on generative models and inverse problems in medical image analysis contexts.

Research overview

I’m broadly interested in questions that can be answered by constructing new imaging datasets, real or otherwise. For example: can we predict invariant nonlinear representations for any image without any dataset-specific training by training on randomized image synthesizers? (Yes. See our ICLR 2025 paper where we obtain state-of-the-art 3D registration and few-shot segmentation using this approach)

Other topics I actively work on include estimating atlases of image collections, predicting image deformations, and constructing equivariant networks for geometrically-structured images.

Recent updates