Rohun Agrawal

I am a Computer Science PhD student at Columbia co-advised by Micah Goldblum and Pavel Izmailov. I am also a visiting researcher at NYU CILVR. My doctoral work is supported by the NSF Graduate Research Fellowship.

Previously, I completed my B.S. in Applied and Computational Mathematics from Caltech where I worked with Georgia Gkioxari on visual reasoning and in Katie Bouman’s lab on imaging inverse problems. I have also interned at Apple training large video models, and at the NASA Jet Propulsion Laboratory calibrating GPR models.

I am interested in developing methods to push the temporal horizons of machine learning models. Recently, I have approached this with world model planning for efficient future horizons, and sparse attention retrieval for longer past horizons. I believe that improving both these directions of horizons will allow agents to be truly persistent and capable.