Pranay Jain
LightNews LightNews
pranayj.bsky.social
Pranay Jain
@pranayj.bsky.social
69 followers 480 following 3 posts
Machine Learning Researcher @ Mitre
Medical Imaging Enthusiast 🩻
The Dellas 🎸
Go birds 🦅
Washington DC 📍
Posts Replies Media Videos
Reposted by Pranay Jain
pfrazee.com
Hey everybody! You may notice Trending start to appear in your app!

Here’s some quick info:

- This is a beta
- Search page on mobile, right sidebar on desktop
- You can disable it with the X button on the trending section
- English only for now, while we test it

Merry Christmas!
Trending BETA What people are posting about. Sixers A Complete Unknown King Henry Joel Embiid Pentatonix Jayson Tatum Jrue Holiday Scrooge Travis Kelce Mavs Mikal Bridges
December 25, 2024 at 10:59 PM
Reposted by Pranay Jain
arnauddoucet.bsky.social
The slides of my NeurIPS lecture "From Diffusion Models to Schrödinger Bridges - Generative Modeling meets Optimal Transport" can be found here
drive.google.com/file/d/1eLa3...
BreimanLectureNeurIPS2024_Doucet.pdf
drive.google.com
December 15, 2024 at 6:40 PM
Reposted by Pranay Jain
ruiqigao.bsky.social
A common question nowadays: Which is better, diffusion or flow matching? 🤔

Our answer: They’re two sides of the same coin. We wrote a blog post to show how diffusion models and Gaussian flow matching are equivalent. That’s great: It means you can use them interchangeably.
December 2, 2024 at 6:45 PM
Reposted by Pranay Jain
lucamb.bsky.social
Nuggets of generative diffusion
November 22, 2024 at 9:30 AM
Reposted by Pranay Jain
theathletic.com
Rookie Jared McCain led the 76ers to their third win of the season.

◽️ 30 PTS
◽️ 6 REB
◽️ 11/20 FG
◽️ 6/11 3-PT FG

It's his 7th consecutive game with 20+ points.
November 23, 2024 at 2:38 AM
Reposted by Pranay Jain
hankgreen.bsky.social
I wish I could tell trees that the universe exists...
November 19, 2024 at 7:30 PM
pranayj.bsky.social
Been working on synthetic medical image generation using diffusion models and trying to convince myself of the benefit of synthetic images, either for privacy purposes, model training or something else. Anyone have insights to share?
#diffusion#machinelearning#medicalimaging
November 19, 2024 at 7:30 PM
Reposted by Pranay Jain
gdalle.bsky.social
The abstract alone is a banger. Also note the detailed experiments section. I have to ask the authors: what comes next?
Abstract A curious defect of the Hamiltonian Monte Carlo algorithm, an otherwise state-of-the-art Markov chain Monte Carlo scheme for simulating from an unnormalized density, is that despite its name, it has very little to do with Lin-Manuel Miranda’s hit musical Hamilton. To remedy this flaw, we propose Alexander Hamiltonian Monte Carlo, a revolutionary new variant of HMC in which the Metropolis-Hastings accept/reject step is replaced by an elaborate duel between the proposed position and the previous state of the chain. The result is a loud (in the sense of noise), diverse (in the sense of variance) celebration of both Bayes and Broadway, which is guaranteed to converge to the room where it happens, if the user is willing to wait for it. 3. Experiments The experiment lasted two minutes, maybe three minutes, and everything it said was in total agreement. 4. Practical Advice on the Application of AHMC 1. It may be tempting to read every sample it writes you, searching and scanning for answers in every line, for some kind of sign. However, we caution users against this behavior (“watching it burn in”); early samples are best erased from the narrative. 2. Users in New Jersey may find it useful to replace the AHMC kernel with any stochastic procedure for proposing and then accepting a new state x 0. (Everything is legal in New Jersey.) 3. Unlike sequential Monte Carlo and other algorithms based on importance sampling, Alexander Hamiltonian Monte Carlo unfortunately does not yield estimates of the MLL (marginal log likelihood), at least not out of the box. However, one can use the output of AHMC to obtain unbiased estimates of the LMM (Lin-Manuel Miranda), as depicted in Figure 1. We emphasize the novelty here: most existing estimates of LMM’s value are biased. 5. Related Work (by blood) Unimportant, there’s a million things I haven’t done, just you wait, just you wait. 6. Discussion To be honest we are doubtful AHMC will prove useful to practitioners. But you never know. After all, who can say which ideas live? Which die? Who cites our story?
November 19, 2024 at 6:22 AM
Reposted by Pranay Jain
lyle.wrathclub.com
Let's be real the only reason anyone watched that Jake Paul fight was for the 1% chance that Mike Tyson would bite something off him
November 17, 2024 at 11:39 AM