We are excited to release the weights of @msftresearch.bsky.social's COLIPRI, our 3D vision–language encoder for chest CT scans, on @hf.co 🤗
Model: aka.ms/colipri
Demo: aka.ms/colipri-demo
Paper: aka.ms/colipri-paper
Why does COLIPRI matter?
🧵0/12 👇
🩻Excited to share our latest preprint: “Data Scaling Laws for Radiology Foundation Models”
Foundation vision encoders like CLIP and DINOv2 have transformed general computer vision, but what happens when we scale them for medical imaging?
📄 Read the full preprint here: arxiv.org/abs/2509.12818
The world’s first multimodal, bilingual radiology dataset could reshape the way radiologists and AI systems make sense of X-rays. PadChest-GR, developed by the University of Alicante with Microsoft Research, has the potential to advance research across the field for years to come. msft.it/6013SLDYZ
"Socialism”
Professor and ‘Godfather of AI’ Geoffrey Hinton says that socialist policies are the only way to ensure that wealth generated from AI is fairly distributed
#Peston
Happy to announce a collaboration with the Mayo Clinic to advance our research in radiology report generation!
newsnetwork.mayoclinic.org/discussion/m...
Tagging some of the core team: @valesalvatelli.bsky.social @fepegar.com @maxilse.bsky.social @sambondtaylor.bsky.social @anton-sc.bsky.social
Have you ever wondered how to train an autoregressive generative transformer on text and raw pixels, without a pretrained visual tokenizer (e.g. VQ-VAE)?
We have been pondering this during summer and developed a new model: JetFormer 🌊🤖
arxiv.org/abs/2411.19722
A thread 👇
1/
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.
Hey folks, we have a final request for comments for our JuliaHealth Mini-Symposium draft submission! 🧪 🛟🩺
Have any final thoughts? Click on the link to learn more! We will be submitting December 1st!
Let's bring #julialang to health research, together!
#medsky #AcademicTwitter #publicHealth
I'm not sure I agree with this take. Good news shouldn't be restricted to only those who can afford it. That's why if you can afford it, it's important to back independent news organisations which are run off donations and declare how they are funded.
BayesFlow is a library for amortized Bayesian inference with neural networks.
⋅ Multi-backend via Keras 3: Use PyTorch, TensorFlow, or JAX.
⋅ Modern nets: Flow matching, diffusion, consistency models, normalizing flows, transformers
⋅ Built-in diagnostics and plotting
🔗 github.com/bayesflow-or...
Figure of the REPA article depicting the distillation architecture and how it speeds up training.
This figure shows that the FID score decreases (better) when the visual encoder used for distillation has higher validation accuracies.
We know that diffusion models learn interesting representations during training, but can we increase their generation performances by distilling representations learned by self-supervised encoders? 👀
The REPA article by Sihyun Yu et al. shows better FID + faster convergence arxiv.org/abs/2410.06940
I am utterly delighted by this subversive use of the Bluesky label feature, where you can subscribe to a custom app that then adds visible labels to profiles
This one shows you which posh expensive private school British public figures attended, plus their current fees
bsky.app/profile/dadd...
I’ve created a starter pack! Follow us if you want to know more about what we’re up to at Microsoft Research Health Futures 🩻⚕️🏥
Get in touch if you think you or someone else should be added!
We have to stop treating planet as a dumping ground.
"Generative AI applications alone could add 1.2 million to 5 million metric tons of this hazardous trash by 2030."
#PlanetaryBoundaries #PublicHealth
www.scientificamerican.com/article/gene...
