Amit Pal Singh Kohli
About me
My name is Amit Kohli, I am a postdoctoral researcher at UC Berkeley coadvised by Professor Matei Zaharia and Professor Meng-meng Fu.
I am working on developing novel therapies for neurological diseases, with a particular focus on
leveraging recent advancements in AI-based structural prediction to better target pathogenic molecules
in gliopathies. I am also exploring how modern AI retrieval tools can be applied to consolidate
biomedical knowledge in a reliable and accessible manner.
I completed my PhD at UC Berkeley with Professor Laura Waller,
where I worked primarily in the field
of computational imaging with a particular interest at the intersection of optics and statistics.
The goal of my PhD research was to design and analyze robust
computational imaging systems—ones
which enjoy algorithmic enhancements but remain impervious to the inevitability of system errors and
imperfect models.
Outside of research I enjoy listening to and making music; I play guitar and bass, and have a passion
for producing and post-processing recorded music. I also
love to be active; I dance Bhangra and play sports
including basketball, baseball, football, tennis, and golf, to name a few. I especially enjoy going to
watch the local Bay Area sports teams in-person! Lastly,
in my spare time, I love to hang out with my family, including my brother's dog, Buster.
Fields of interests: Molecular biology, Glia, Optics, Statistics, Computational Imaging, Signal Processing, Machine Learning/AI.
Contact me
-
I am best reached via email: apkohli at berkeley dot edu.
Recent news
- January 2026 - I'm excited to start my postdoc at UC Berkeley coadvised by Prof. Matei Zaharia and Prof. Meng-meng Fu, where I'll be working on developing novel therapies for neurological diseases.
- March 2025 - Ring deconvolution microscopy: exploiting symmetry for efficient spatially varying aberration correction was just published at Nature Methods 2025. Check out the accompanying codebase: rdmpy
- January 2025 - Our paper, Noise2Image: noise-enabled static scene recovery for event cameras is now on published in Optica. This paper introduces a method for recovering static scenes from event camera data using shot noise!
- February 2024 - Our paper, Wavefront Randomization Improves Deconvolution is now on arXiv. Stay tuned for a publication.
- November 2023 - Our paper, Random optics for improved digital aberration correction, was just accepted to Photonics West 2024 for an oral presentation. This paper demonstrates a simple way to make an imaging system entirely aberration-free!! Stay tuned for the publication.
- August 2023 - Introducing Ring Deconvolution Microscopy: An Exact Solution for Spatially-Varying Aberration Correction. This exciting new paper provides a easy-to-use method for deblurring images that have spatially-varying blur. We also released an accompanying python package called rdmpy. This paper introduces a new image deblurring method, which incorporates physical information about the imaging system.
- January 2023 - We presented SeidelNet: an aberration-informed deep learning model for spatially varying deblurring at Photonics West 2023. This paper introduces a new image deblurring method, which incorporates physical information about the imaging system.
- July 2022 - We presented Image-to-Image Regression with Distribution-Free Uncertainty Quantification and Applications in Imaging at ICML 2022. This paper develops a method for adding for distribution agnostic, per-pixel prediction intervals to the output of any image-to-image regression model. Check out the poster and codebase.
- June 2022 - "Linear Revolution-Invariance: Modeling and Deblurring Spatially-Varying Imaging Systems" is up on arXiv. This paper develops and analyzes an exact forward model for rotationally symmetric imaging systems; inverting it produces a fast, accurate deblurring method. Try it out yourself with the accompanying codebase!
- May 2022 - Rigorous Uncertainty Estimation for MRI Reconstruction was presented at ISMRM 2022. I had a blast collaborating with Ke Wang in Prof. Miki Lustig's group for this one! The paper applies conformal prediction to MRI reconstructions of undersampled data in order to get rigorous, image-valued prediction intervals.