In Vancouver this week for #ICML2025 ! If you would like to chat over a cup of coffee about LLM alignment, Adaptation, or any other topic, feel free to reach out!
Our workshop is bringing together top-tier researchers, groundbreaking ideas, and an incredible speaker lineup. Let's talk about robustness, shifts, and adaptation! Be there!๐ฅ #icml2025@icmlconf
๐ฃ๏ธ Kate Saenko, Xiaoxiao Li, Deepak Pathak, Rahaf Aljundi, and Gao Huang
Ever noticed how one scene seamlessly transitions into another in films? Thatโs a match-cutโa subtle yet powerful cinematic trick.
Our MatchDiffusion, generates two videos from text prompts, designed to form a seamless match-cutโeffortlessly and training-free. ๐ฅโจ 1/n
๐Huge congratulations to Motasem Alfarra @MotassimF from the @KaustVision IVUL family on successfully defending his PhD dissertation, โCombating Domain Shifts in the Deep Learning Era,โ under the guidance of Prof. @BernardSGhanem๐
๏ฟฝ๏ฟฝWishing you all the best in your next steps!
Excited to be in Vienna ๐ฆ๐น for #icml2024!
If you are there, let's grab a coffee and talk about domain shifts, adaptation and trustworthy AI.
P.S, I am presenting a paper about realistic evaluation of test-time adaptation methods.
Preprint: https://t.co/qo55C85o8b
Checkout our recent #MICCAI 2024 paper about evaluating distribution shifts in medical imaging! Huge kodos to the talented first authors @KumailAlhamoud@YasirGhunaim and all the mentors.
๐ Happy to share my first paper as a PhD student with @MarzyehGhassemi! It's accepted to MICCAI 2024! Hereโs a thread:
1/ ๐ Introducing FedMedICL: A framework for evaluating federated continual learning in medical imaging, capturing label, demographic, and temporal shifts.
Heading to Seattle for #CVPR2024! ๐โ๏ธ I will be helping organizing the first workshop on Test-time Adaptation @MAThyself !
If you are interested in Test-time Adaptation or continual learning, i am very happy to connect over a cup of โ๏ธ!
Can a language model understand JPEGs ๐พ? We trained ours ๐ฆ on raw byte streams, and it learned to speak JPEG-ese! We find the model's can understand file properties, fix anomalies, and generate new JPEGs ๐ผ๏ธ, all while bypassing the decompression step ๐.
A thread ๐งต
๐จ Heads up for our workshop day! ๐จ We've got a stellar lineup of speakers. Our Speaker of the Day is Alexei Efros, professor at UC Berkeley's EECS Department. His research in test-time adaptation has profoundly shaped the field!
ูุฌูุฏู ุฎุงุฑุฌ ุญุฏููุฉ ุงูุญููุงู ูุง ูุนูู ุนุฏู ุงูุชู ุงุฆู ููุง
Your existence outside the zoo does not mean you do not belong there