←
Ehsan Adeli1,144 posts
Ehsan Adeli avatar

Ehsan Adeli

@eadeli

Prof. at @Stanford @StanfordMed @StanfordPSY; @StanfordSVL; @StanfordAILab; @StanfordTAILab Computer Vision, Neuroscience, Healthcare, Medical Image Analysis.

◷ Joined July 2009▧ 59 Photos and videos
Ehsan Adeli avatar
Ehsan Adeli@eadeli·Nov 13

A day without work email. #OutlookOutage It kinda felt good. I was only connected with my own team, over Slack, in-person, and on Zoom! We got a lot of work done.

Ehsan Adeli avatar
Ehsan Adeli@eadeli·Nov 12

One of my #ICLR submissions got three reviews with almost identical concerns and questions, just worded a little differently. Either the reviewers were remarkably in sync, or their LLMs were. Same service? Same model version? Welcome to the AI peer review era. 😄

Ehsan Adeli avatar
Ehsan Adeli@eadeli·Nov 11

How should you tokenize the brain? Follow its hierarchy. 🧠 Our #NeurIPS2026 paper introduces HBAR, which turns voxel-level resting-state fMRI into compact discrete tokens at three scales: voxel groups → parcels → functional networks. - Up to 415× compression while keeping sub-parcel detail that standard parcellations average away - Better parcel- and network-level reconstruction than rate-matched flat tokenizers (HCP-YA) - Structure matters: shuffle the anatomical grouping and parcel correlation falls from 0.920 to 0.886 - Atlas-based, fully learned, or soft prior-guided hierarchies, so we can test what brain priors actually buy you A practical building block for brain foundation models. Led by @karanps, with @__yuezhao__ , and Mohammad Abbasi. 📄 arxiv.org/abs/2611.11081 🌐 karanps.com/hbar @stanfordtailab @StanfordAILab @StanfordSVL @NeurIPSConf

What should a token look like when your data isn’t text or pixels? 🧠 HBAR is now on arXiv 🚀 arxiv.org/abs/2611.11081 We explored this question with fMRI, using the brain’s known multi-scale organization to build a hierarchical tokenizer. Accepted to NeurIPS 2026.

Ehsan Adeli avatar
Ehsan Adeli@eadeli·Nov 9

Reconstruct the person behind the camera and the world in front of it. See our latest work, EgoMo3R: full-body motion + 3D scenes from first-person video. Motion guides geometry. Geometry grounds motion. @chaitanya100100 will present the work at #NeurIPS2026 in December.

Can a human motion prior help egocentric SLAM, and can SLAM in turn help egocentric motion reconstruction? Presenting EgoMo3R, our #NeurIPS2026 paper on joint 3D motion and scene reconstruction from egocentric video. chaitanya100100.github.io/Eg…

0:08
Stanford HAI avatar
⟳ Ehsan Adeli retweeted
Stanford HAI@StanfordHAI·Nov 3

Are you developing AI or technology to support older adults, including people living with Alzheimer's disease and related dementias? Stanford AITC has launched a research award competition, supported by the National Institute on Aging. Join a webinar to learn more: November 4 | 9 AM PT / 12 PM ET: stanford.zoom.us/webinar/reg… (1/2)

Ehsan Adeli avatar
Ehsan Adeli@eadeli·Nov 2

So great to be part of the Stanford Digital Brain Project! With @dyamins and an amazing team at @StanfordBrain, we're building a digital model of the human brain to learn how it works and, one day, better understand and treat brain and mental health disorders.

Introducing the Stanford #DigitalBrain Project! This new initiative will use advanced brain imaging and machine learning to build predictive models of human brain activity and lay the groundwork for more personalized treatments for brain disorders. 🔗 neuroscience.stanford.edu/ne…

Post attachmentPost attachmentPost attachment
Ehsan Adeli avatar
Ehsan Adeli@eadeli·Oct 19

Patient activity in the hospital gets scattered across shifts, notes, and handoffs. Our recent work, TRACER, pulls it back together: a training-free harness that lets LLMs/VLMs track activity from bedside video + EHR. Harness and synthetic data are open. Led by @DuranteZane 👇

1/ Can AI help doctors make more informed decisions based on patient activity and behavior over the course of a hospital stay? Today we introduce TRACER: a training-free multimodal harness that enables LLMs and VLMs to track patient activity using bedside video and health records. Blogpost: ai.stanford.edu/blog/tracer/ Paper: arxiv.org/abs/2610.21266

0:42
Ehsan Adeli avatar
Ehsan Adeli@eadeli·Oct 17

It feels like déjà vu to trade one homeland for another, only to end up in a kakistocracy!

Ehsan Adeli avatar
Ehsan Adeli@eadeli·Oct 16

Do text-to-image models need language pretraining? We trained Halcyon from scratch on 100M image–text pairs with no LLM at all, and it beats LlamaGen & JiT-T2I using 18–275× less compute. Open weights, 0.1–4B, in recent work led by @keshigeyan; @__yuezhao__ @jcniebles @drfeifei.

Excited to share Halcyon, our latest research on multimodal pretraining! Building strong text-to-image generative models is simpler than we think! Blog: radicalnumerics.ai/blog/halc… Models: huggingface.co/collections/r… Paper: radicalnumerics.ai/assets/ha…

Yaser Zhian avatar
⟳ Ehsan Adeli retweeted
Yaser Zhian@yaserzt·Oct 15

Maybe I'm dense, but it only just occurred to me that this "AI revolution" is the only pure-software technology jump that I can think of. Wheels, steam engine, cars, electricity, cinema and TV, computers, the internet, mobile phones, etc. were all enabled by new hardware.

Ehsan Adeli avatar
Ehsan Adeli@eadeli·Oct 9

It is so funny how the tech industry spent 20 years building tests to prove you're not a bot. Now we're building AI agents to pass them for us to reserve a table online. We're now engineering the exact thing we spent years engineering against. Same tech, opposite team.

Karan Singh avatar
⟳ Ehsan Adeli retweeted
Karan Singh@karanps·Sep 24

LLMs have tokens. Images have patches. What should the brain have? 🤔 HBAR, our hierarchical tokenizer for fMRI, was just accepted to NeurIPS 2026! 🎉 It compresses fMRI ~415× while preserving structure from voxels → parcels → functional networks.

Post attachment
Puyin Li avatar
⟳ Ehsan Adeli retweeted
Puyin Li@PuyinLi·Sep 23

💰 Announcing prizes for the QuantiPhy Challenge @ NeurIPS 2026! 🏆 $3,500 total cash prizes Main Track 🥇 $1,000 · 🥈 $500 · 🥉 $250 Open-Weight Track 🥇 $1,000 · 🥈 $500 · 🥉 $250 ⏰ Team registration: Oct 9, 23:59 AoE ⏰ Submission deadline: Oct 23, 23:59 AoE

Tiange Xiang avatar
⟳ Ehsan Adeli retweeted
Tiange Xiang@xxtiange·Sep 24

Today we announce 💰exciting prizes💰 for winners at the QuantiPhy Challenge @NeurIPSConf 2026. Register your team before 9th Oct. and submit csv files before 23rd Oct. to join the NeurIPS competition and win the prize!!! See more at: quantiphy.stanford.edu/compe…

💰 Announcing prizes for the QuantiPhy Challenge @ NeurIPS 2026! 🏆 $3,500 total cash prizes Main Track 🥇 $1,000 · 🥈 $500 · 🥉 $250 Open-Weight Track 🥇 $1,000 · 🥈 $500 · 🥉 $250 ⏰ Team registration: Oct 9, 23:59 AoE ⏰ Submission deadline: Oct 23, 23:59 AoE