I’ve launched 2 new course websites for my Introduction to #ComputerVision and #DeepLearning in Computer Vision courses.
🆕 Gaussian Splatting
🆕 Flow Matching
🆕 Sparse Autoencoders
🆕 Lecture Notes (work in progress)
The videos don’t include voiceovers, but that’s coming in a future revision 🤞
Share your feedback in the thread.
Latent diffusion needs no decoder😵
New paper: DWD, a linear head on DiT predicts pixels directly
• Same or better quality on every tested latent model, less flicker in Wan video
• Even pixel models improve: ImageNet FID 1.55 ➡️ 1.43 (pixel SOTA), photorealistic texture in T2I
Still refreshing OpenReview, waiting for that one reviewer to reply? 🫠
So I checked how much the #ICLR2027 rebuttal has actually moved scores so far 📊
Nov 6 → Nov 16, 12:26 UTC · ~40k papers
• Only 9% of papers changed score, and ups beat downs 12 to 1
• 64% of all review changes were 2→3. 1→4 happened just 4 times
• Papers at 2.5–2.75 moved up most (13–19%). At 1.0: 0.3%
• 1 in 4 papers that started at 1.0 has withdrawn
I'll keep updating this until the discussion closes on Nov 18. Look up your own score here 👇
juevn.github.io/iclr2027-rev…
Vanderbilt is hiring 52 (!) faculty in our new College of Connected Computing. I'm hoping some philosophers will be among them. See especially the Computing & Society Initiative (computing.vanderbilt.edu/com…) which could be an excellent fit
We're proud to announce the launch of CS Learn, a new online learning platform that offers free access to select courses from Carnegie Mellon University’s top-ranked computer science program.
Check out the course offerings: bit.ly/4C4770z
I will never understand why you wouldn’t split the paper across multiple .tex files 🤓 Keeps content organized. If you use Overleaf, you are one double click away from the desired content.
Introducing Orion, a foundation model that predicts where you look. Orion finds twice as many UX bugs as frontier models, cuts computer-use latency by 31%, and even won a drone race.
Video models can generate scenes. Geometry models can reconstruct them.
What if one model did both?
UGG generates video, depth and camera motion together, as one scene. (NeurIPS 2026)
Co-first authors: @ZhiyiLi02 and @ruojin8
🌐 ugg-model.github.io
A few years have passed since our AI-powered drone, SWIFT, challenged the world-champion drone racers. Watch the full story behind that historical milestone for physical AI in this 25-minute documentary by director Valerio Jalongo! Meet the people and discover the science behind that research! Now on YouTube: youtu.be/6kpvOAsq8o0
A big thank you to the people behind it: Elia Kaufmann, Leonard Bauersfeld, Antonio Loquercio, Matthias Mueller, Vladlen Koltun, Christian Pfeiffer, Thomas Laengle, Alexander Barden, Yunlong Song, Drew Hanover, Florian Trautweiler, tamar tolcachier, Rene Ranftl, Mathias Gehrig, Daniel Gehrig, Manasi Muglikar, Chao Ni, Nico Messikommer, Marco Cannici, Adriana Mohap, Gonca Yılmaz, Shreedhar Govil, Thomas Bitmatta, Marvin Schaepper, Alex Vanover
@uzh_ifi@UZH_Science@UZH_en@UZHspacehub@ERC_Research@UZH_ai@AUTOASSESS_EU@UZHspacehub, @ELLISforEurope@captainvanover#FPV#Drone#DroneRacing#FPVRacing
We would like to introduce SaiKiran Tedla as a Cohort 4 #SchmidtAIinSci Fellow!
Supervisors: Profs. Kiriakos Kutulakos & David Lindell.
Project: ‘Generative Re-Photography: Towards Flexible Post-Capture Camera Editing 📷
Can a 2M-parameter network learn bundle adjustment? 🤔
New preprint: DeepBA recovers poses, focal lengths and depth for any number of views, starting from a trivial cold start. 🧵👇
📄 arxiv.org/abs/2611.18256
🌐 cphjsn.github.io/deep-ba
For this paper I wrote and released a tex->html script so I could self-host an HTML version of the paper that matches my homepage style. If you're already using my homepage template you should try it: github.com/jonbarron/jonbarr…
Hi all, I have a new result I'm excited to share. It turns out that *all existing implementations* of softmax and cross-entropy (that's JAX, PyTorch, textbooks, etc) have a subtle bug --- two bugs, actually. A simple rewrite fixes both, and eliminates "slingshots" when grokking: