Can VLMs help robots perform complex multi-stage tasks in zero shot? MOKA integrates VLM-based reasoning at multiple stages in a hierarchy to get robots to get zero-shot skills, and further refines those skills through experience: moka-manipulation.github.io Thread 👇

Is it just me or do you guys agree 20s should be measured in dog years? Like only about 5 years went by but I already feel 30 years older...

Our survey "Hyperbolic Deep Learning in Computer Vision: A Survey" has been accepted to #IJCV! The survey provides an organization of supervised and unsupervised hyperbolic literature. Online now: link.springer.com/article/10.100… w/ Mina Ghadimi Martin Keller-Ressel jeff @syeung10

This is hilarious, open thread at your own risk 😂

github.com/google-researc… Yes. Yes!!!! Everyone read this material three times!

Me waiting for the #ECCV2024 reviews release

Already missing the sun 🥹

🚀 Just launched a new repo compiling key papers on Explainable CV! 🔍 I've added short summary for each paper, with a major focus on concept-based explanations—I'm expanding to include other directions too! Contributions are welcome! Check it out here: github.com/aisagarw/aweso…

All vision-language models should have hyperbolic embeddings. Vision and language are incredibly hierarchical in nature! See below our latest work on hyperbolic vision-language models that exploit visual compositions through entailment:

I’m excited to share our latest work on generative models for materials called FlowLLM. FlowLLM combines Large Language Models and Riemannian Flow Matching in a simple, yet surprisingly effective way for generating materials. arxiv.org/abs/2410.23405 Benjamin Kurt Miller Ricky T. Q. Chen Brandon Wood

Yesterday, Erik Bekkers presented his vision on equivariance to Informatics Institute University of Amsterdam, showcasing recent work on geometry-grounded representation learning - addressing fundamental limitations in geometric reasoning of current AI systems 🤖 Exciting times ahead for geometric deep learning!

How can AI better handle uncertainty and scale to complex tasks? 🤔 Our research explores hyperbolic embeddings in vision-language models (VLMs), helping us measure how certain a model is about its predictions—something Euclidean embeddings struggle with. 🚀 Here's how 👇

🔍 Our latest blog features the HAVA-Lab, UvA Amsterdam’s first interdisciplinary lab involving all its 7 faculties, highlighting the significance of collaboration in addressing complex societal challenges ✨ 🔗 Read the blog here: ivi.fnwi.uva.nl/ellis/news-/ha…

Hyperbolic learning is growing rapidly by the day. From weekly alerts in 2023 to daily digests in 2024! From our current research, it is clear that 2025 will be a huge year for hyperbolic learning research. I had an interview to elaborate our research: shorturl.at/CQD53

o3-mini is wild. Full version o3 isn't even released yet and people are already doing incredible things. 10 examples:

If you are numbed by image to 3d results, this is generating the skeleton for any mesh so you can animate it, use it in simulation. If you compare how things have advanced in the past 5 years, we used to be able to do only human, hand, animal 3d poses separately . Now this is

A couple of months ago, I told you vision-language = hyperbolic. #ICLR2025 agrees and has made our latest paper an oral presentation! To present at ICLR: Compositional Entailment Learning for Hyperbolic Vision-Language Models. Link: arxiv.org/abs/2410.06912

We will be back at #CVPR2025 with another piece of evidence for "All Vision-Language Models should be Hyperbolic". This time, we show how hyperbolic CLIP makes safety awareness possible! Check out the original post below"

#ICLR2025 was a blast, great to talk with all of you about our works on hyperbolic vision-language models, a better way to do object detection, and brain-aligned image generation! Next destination: the Web Conference in Australia, happening now

Want to go back to these days honestly!