INSID3 segments objects across domains using ONLY ONE annotated example
it works entirely without a segmentation decoder, task-specific fine-tuning, or external mask generators like SAM
CVPR 2026 paper with enormous practical potential
A big takeaway from Day 1 at the #AIResearchSummit: The biggest risk right now isn't adoption, it's inaction. Time to embrace the tool or get left behind. π¨
The era of treating AI as an optional add-on is officially over. #MayoClinic
The energy is high at Mayo Clinic's AI Research Summit as researchers, clinicians, students and industry collaborators come together to explore the future of healthcare AI.
Today's conversations are focused on advancing AI research, fostering new collaborations and exploring how emerging technologies can responsibly drive discovery and impact patient care.
#AIResearchSummit #MayoClinic #AI #HealthcareAI
Today, weβre excited to introduce Miso One, the most emotive voice model in the world.
Miso One is an 8-billion-parameter text-to-speech model for highly expressive speech generation. It emotes like a human and responds faster than a human, with just 110 milliseconds of latency.
Weβve open-sourced the model weights, with API access coming soon.
Hear how Miso One sounds in the thread below.
π A record $50,000 in prizes @MICCAI_Society
The ORena SAVE FOCUS Challenge is one of MICCAI's 2026 lighthouse challenges
Can #AI track foreign objects across an *entire* operation to prevent retained foreign objects?
#MICCAI2026#SurgicalAI#MedicalImaging#AIforHealth
@elonmusk@SawyerMerritt@Starlink Flew Delta recently after experiencing Starlink on other airlines, it genuinely felt like a step back. It was painful.
That's true, but isn't switching to a single 'router' provider conceptually similar?
If they go down for whatever reason, your project will suffer. Still need a fallback off that router to multiple routers/providers if one goes offline. Then paradoxically - in that case, I have created my own router and don't need them?
Catching up on a great few weeks in #SurgEd. Presented our AI and eye-tracking work at @MayoClinicSurg. Incredible to share the room with two people pushing the boundaries of AI, @Laparoscopes and @_TylerLoftus.
β Now, gearing up for #SEW2026. The conversations don't stop!
Blessed to have matched at my top and favorite program, University of Wisconsin! @uw_medicine.
Thank you to all my family and mentors who made this possible! #Match2026
Introducing π¨ππππππππ πΉππππ ππππ: Rethinking depth-wise aggregation.
Residual connections have long relied on fixed, uniform accumulation. Inspired by the duality of time and depth, we introduce Attention Residuals, replacing standard depth-wise recurrence with learned, input-dependent attention over preceding layers.
πΉ Enables networks to selectively retrieve past representations, naturally mitigating dilution and hidden-state growth.
πΉ Introduces Block AttnRes, partitioning layers into compressed blocks to make cross-layer attention practical at scale.
πΉ Serves as an efficient drop-in replacement, demonstrating a 1.25x compute advantage with negligible (<2%) inference latency overhead.
πΉ Validated on the Kimi Linear architecture (48B total, 3B activated parameters), delivering consistent downstream performance gains.
πFull report:
https://t.co/u3EHICG05h
Here is our updated database of grants for early careers researchers in all fields.
It goes way beyond traditional NIH and NSF funding opportunities. We list 428 types of grants.
Download it here: https://t.co/RGXGrdp2uf
For the 8th straight year, Newsweek has named Mayo Clinic the No. 1 hospital in its Worldβs Best Hospitals rankings.
"Finding more cures for patients is what drives our staff to transform healthcare so that new discoveries can benefit people everywhere," says CEO @GFarrugiaMD.
Read more: https://t.co/SQ2MHaikY7
The assumption: train longer, see better.
The reality: some junior residents already exhibit expert-like gaze patterns. Some seniors still scan like novices.
Gaze reflects cognition.
Where you look, how long you fixate, how you scan, these aren't random. They're windows into how you process surgical information. #ASC2026
We assess surgical competency by time in training. But "expert" isn't binary, neither is "novice."
Eye-tracking reveals a spectrum of visual strategies within every PGY level. Two PGY3s can watch the same operation and see it completely differently.
Phenomenal AAS Presidential address by Dr. @BrenessaL this morning.
The visual that resonated with me was the interwoven threads.
Education is a noble and vital mission. Without it, academia ceases to exist.
#ASC2026@AcademicSurgery
Closing 2025 grateful for progress in deep learning, computer vision, and eye-tracking.
..and best of all, learning to be a dad πΆ! The ultimate reminder that the best systems are built with care, tested with patience, and improved with love. Excited for 2026!
π Happy New Year!