Vatsal Baherwani

Hello, welcome to my website!

About Me

I am a PhD student at New York University, advised by Pavel Izmailov and Andrew Gordon Wilson.

I previously graduated with a bachelor's degree from the University of Maryland, where I was fortunate to work with Abhinav Shrivastava, Ashwinee Panda, and Tom Goldstein.

I occasionally post essays on my blog. I use this as an exercise to write more often and get used to publishing imperfect work publicly, so I would be happy to hear feedback on any of my posts.

Recent life updates:

09/25 Starting my PhD at NYU!

08/25 Finishing my internship at the Center for Human-Compatible AI at UC Berkeley. Very grateful to have worked with Raj Movva and Emma Pierson, from whom I learned so much about practical applications for ML interpretability.

05/25 Saying goodbye to UMD :( Thank you to all of my mentors, especially Dr. Pete Kyle who took a chance on me and first introduced me to academic research.

Email  |  CV  |  Google Scholar  |  Twitter  |  GitHub  |  LinkedIn

Publications

We develop a guardian model that can dynamically adapt to ensure an LLM complies with arbitrary user-specified rules.

We approximate the dense backward pass of a sparse mixture-of-experts model, leading to improved training stability and performance with negligible overhead.

We show that motion information is independently learned in early timesteps of the diffusion process, prior to the materialization of spatial attributes. We use this insight to present a simple and efficient method for targeted video motion customization.

This website is built on a template by Jon Barron.