Amit Kohli

PhD student at @Berkeley_EECS supervised by Laura Waller (@optrickster). Computational aberration correction. Statistical imaging. Undergrad at @stanford_ee.

[Anastasios Nikolas Angelopoulos @ml_angelopoulos]

After many months of development, LMArena has transitioned to a new user experience.

It's way better, check it out, and congrats to the team that made it possible. :)

[Dekel Galor @dekelgalor]

Check out our new preprint! Very fun project co-led with the amazing @hadivafaii

[Amit Kohli @the_legitamit]

Github: https://t.co/Zreltym4jj

The package will be pip installable starting next week 😎

Big thanks to my coauthors @ml_angelopoulos, @davidrmcall, Esther Whang, @Kyrollosyanny, @sixian_you, @FedeGasparoli, @bojuichang, @RetoPaul, Laura Waller.

[Amit Kohli @the_legitamit]

Ring deconvolution microscopy is now published at @naturemethods!

https://t.co/UWTo4Es9Gj

There are some fun new additions including light-sheet deconvolution 🫡

Stay tuned for the official python package release next week! Any feature suggestions are more than welcome 😃

[Amit Kohli @the_legitamit]

🕺🏽Introducing a new deblurring pipeline:

💍Ring Deconvolution Microscopy (RDM)🔬

https://t.co/LtKBQX1WCt

RDM enables exact spatially-varying deblurring with one calibration image.🤯

It’s significantly better than standard deconvolution but just as easy to use!💪

A 🧵 (1/6)

[Anastasios Nikolas Angelopoulos @ml_angelopoulos]

🚨 New Textbook on Conformal Prediction 🚨

https://t.co/n2gcyiOM7k

“The goal of this book is to teach the reader about the fundamental technical arguments that arise when researching conformal prediction and related questions in distribution-free inference.

Many of these proof strategies, especially the more recent ones, are scattered among research papers, making it difficult for researchers to understand where to look, which results are important, and how exactly the proofs work.

We hope to bridge this gap by curating what we believe to be some of the most important results in the literature and presenting their proofs in a unified language, with illustrations, and with an eye towards pedagogy.”

We are looking for feedback — and this is only a draft, with Part 4 coming soon! Please reach out!

With Rina Foygel Barber and @stats_stephen

[Amit Kohli @the_legitamit]

Check out this film @terasajna and I made together!

[Amit Kohli @the_legitamit]

A cool way to get static scene components for free with an event camera!

[Amit Kohli @the_legitamit]

@DocDellaire @ml_angelopoulos @sixian_you @Kyrollosyanny @optrickster Great idea! I agree, I'll give NLM a shot (pun intended 😉). Thanks for engaging with the codebase and for the feedback :)

[Amit Kohli @the_legitamit]

🕺🏽Introducing a new deblurring pipeline:

💍Ring Deconvolution Microscopy (RDM)🔬

https://t.co/LtKBQX1WCt

RDM enables exact spatially-varying deblurring with one calibration image.🤯

It’s significantly better than standard deconvolution but just as easy to use!💪

A 🧵 (1/6)

[Amit Kohli @the_legitamit]

@_jasonliu_ Thanks Jason! It highlights that there are representations of images that look nothing like the image (e.g., fourier transform). We can design imaging systems that output these representations and then we can map to the image. NNs can learn these mappings (though not used here).

[Amit Kohli @the_legitamit]

A new way to correct aberrated imaging systems!

🌊Wavefront Randomization🌊

https://t.co/tPxclPCMTC

Aberrations make images blurry ☹️
Deconvolution helps but is sensitive to noise 😬

BUT if we put a ✨RANDOM✨ mask into the system, deconvolution works better!

A 🧵 (1/5)

[Amit Kohli @the_legitamit]

Big thanks to my coauthors @ml_angelopoulos and @optrickster as well as the rest of Wallerlab ♥️

Shoutout to random masks 🎭

(5/5)

[Amit Kohli @the_legitamit]

Better transfer functions ➡️ better deconvolution!

With no zeros, the randomized transfer function is less sensitive to noise that would usually overcome signal.

The result?

Robust deconvolution whose quality does NOT depend on aberration type/severity 🔥

(4/5)

[Amit Kohli @the_legitamit]

🙏🙏

[IEEE Trans. on Computational Imaging @IEEE_TCI]

Impressive results!

[Amit Kohli @the_legitamit]

@neurotheory Thank you for providing them!!

[Amit Kohli @the_legitamit]

Big thanks to my coauthors @ml_angelopoulos, David McAllister, Esther Whang, @sixian_you, @Kyrollosyanny, @optrickster, and all of Wallerlab ❤️

(6/6)

[Amit Kohli @the_legitamit]

A faster alternative: DeepRD.

DeepRD uses Seidel coeffs to generalize out-of-distribution.

It disentangles (!) and corrects for each aberration independently.

Here we sweep the distortion coefficient. It only undistorts the field without applying any other corrections.

(5/6)