Just presented our @fau.de @taltstidl.bsky.social #ML study at #TCT2025 🌉 demonstrating 👉 accurate & daily refined predictions of neurological outcome after #cardiacarrest based on routinely collected clinical & laboratory parameters is feasible with #TabPFN 🔗 www.tctmd.com/slide/tct-91...
Thomas Altstidl
@taltstidl.bsky.social
Researcher & PhD Candidate · Machine Learning and Data Analytics Lab @fau.de
Just in. Modern #FijiAirways A350 on its inaugural flight into NUE with special @fau.de livery #MovingKnowledge. Daily direct flights from Nuremberg to #Fiji start today, April 1, and #FAU members will receive a special 17.43% discount in homage to our university’s year of birth #1innovator
Happy to have presented our work on a cross-platform smartphone auscultation SDK and optimized filters for severe aortic stenosis detection at this year's #HICSS58. Check out the full paper at hdl.handle.net/10125/109271
Leaving #NeurIPS after two workshops on tabular and time series models, plus short visits to Lynn Canyon and the Vancouver Christmas Market. Shout-out to the @neuripsconf.bsky.social organizers for such a wonderful conference! Who knows, maybe I'll see some of you again in San Diego next year.
In other news: the test of time award went to two papers this year, "Generative Adversarial Nets" and "Sequence to Sequence Learning with Neural Networks." Both of them well deserved 🥳 That's a wrap for the main conference. Now looking forward to some engaging workshops.
At #NeurIPS yesterday we had an invited talk by Danica Kragic illustrating the progress we have made in robotics. And Rosalind Picard reminded us to keep focusing on the human element of our research - in more than one way.
Yes, neural networks can encode information! Since seeing NeRF for the first time I'm always on the lookout for creative ways to use this capability. Here, a #NeurIPS paper using MLPs as signed distance fields for rigid-body collisions: sites.google.com/view/sdf-sim. What else could we use this for?
Yesterday's #NeurIPS invited talk by Lidong Zhou perfectly illustrated how hardware/software co-design has evolved in the age of AI to enable both more efficient and more powerful models. This was followed by a detailed technical review of diffusion models and beyond by Arnaud Doucet.
Stop by East Exhibit Hall A-C #2109 if you're interested in certified robustness at #NeurIPS. I'll be stuck there for three hours so I'm always happy to chat, whether about my work or otherwise 😉
Besides some oral talks and gazillions of posters, yesterday's schedule at #NeurIPS also included invited talks by Sepp Hochreiter, who developed LSTM in his diploma, and Fei-Fei Li, who established the influential ImageNet. Last item on the agenda was the "After Dark" event.
Impressively smooth convergence achieved by Defazio et al. in their paper "The Road Less Scheduled" at #NeurIPS. And since the code is available at github.com/facebookrese... it's easy to try it out on your own machine learning tasks. Time to get rid of the cosine scheduler?
Great day of tutorials and expos at yesterday's #NeurIPS. The tutorial on flow matching was especially enlightening to me. And the best thing is that it's fully open source with a library, guide, and code examples: github.com/facebookrese...
GitHub - facebookresearch/flow_matching: A PyTorch library for implementing flow matching algorithms, featuring continuous and discrete flow matching implementations. It includes practical examples fo...
A PyTorch library for implementing flow matching algorithms, featuring continuous and discrete flow matching implementations. It includes practical examples for both text and image modalities. - fa...
github.com
What better way to meet new friends at #NeurIPS than a hike in the surrounding mountains of Vancouver! Joined a few other conference attendees to see a magnificent waterfall, imposing mountain scenery and a lovely sunset to top it all off.
Flew into Vancouver just in time for a beautiful sunset 🌅 Now looking forward to some sleep and a week of engaging discussions at #NeurIPS
Hard to believe that #NeurIPS is just around the corner. Time to post a sneak peek of my poster. If you're interested in the scalability of certified robustness when adding more and more data, stop by for a quick chat on Thu, Dec 12 @ 11 am - 2 pm in East Exhibit Hall A-C #2109
I've officially joined the X-odus and moved to where the grass is greener and the skies are bluer. From now on I'll be exclusively posting on BlueSky 🦋 Hoping to see old friends, and make lots of new friends here.
What can human vision tell us about computer vision models? If this question interests you, definitely check out these two posters @eccvconf https://twitter.com/dariozanca/status/1840376961358512139
Looks like Friday to Saturday was the best night for the #NorthernLights. Here's a few photos. Not the best, but all I got. It's very rare to see them this far south @erlangen_de and, while faint, it was an amazing experience 🤩
Last Saturday, 200+ students at 8 Bavarian locations competed in #TheTensorTournament hackathon organized by @FAU_MaD_Lab. Congrats to the winners Neural Ninjas, IDEK (both @UniFAU), and Drone Drivers (@TU_Muenchen). Thanks to all organizers and sponsors🎉 https://t.co/vOWOj2V3Tb
Happy to announce "The Tensor Tournament" 2024 🎉 Join us on May 4th @UniFAU and across Bavaria for an unforgettable day of hacking. You can expect amazing prices, plus free food and drinks. Spots are limited, so build a team and secure your spot now 👉 https://the-tensor-tournament.de
The Tensor Tournament
A machine learning hackathon
the-tensor-tournament.de
In the spirit of celestral exploration, I wish all of you a Merry Christmas 🎄 (and if you're not celebrating it, a wonderful time off). Composite X-ray, optical and infrared image shows stars in the NGC 2264 "Christmas Tree Cluster" © NASA and partners https://t.co/JpBk6O55CJ
Huge shout-out to the organizers @k_breininger @VisionBernie @anki_xyz for keeping us productive, entertained and well fed for four days. I had a blast and also stumbled across a few interesting papers at #reproduceCVPR https://twitter.com/k_breininger/status/1734312136132739280
Here's a little known @github feature I just learned about that's great for #scientists. If you add a CITATION file to your repo there will be a quick shortcut to cite your paper in the sidebar. To add your own head to https://t.co/5SagvIbywO
Lots of inspiring talks on #digitalhealth at this year's #ESCCongress! Our poster on smartphone-based phonocardiography for aortic stenosis detection is at http://tinyurl.com/5n7tc95a @jmaltstidl @GaedeLuise @BjoernEskofier @Steph_Achenbach
Detection of aortic stenosis using built-in microphones of commercially available smartphones
Access this presentation on ESC 365 from ESC Congress 2023 on Mobile Apps by Mr J. Altstidl (Germany,DE) on ESC 365.
tinyurl.com
@jmaltstidl presenting our early work on smartphone-based phonocardiography for aortic stenosis detection at #ESC2023 in beautiful #Amsterdam 🌷
With #NeurIPS2023 rebuttals in full swing, who else would like to see @openreviewnet adopt reviewer names similar to @MLflow or @weights_biases run names? I'd very much prefer thanking reviewer industrious-sloth-911 for their thoughtful comments compared to a reviewer GaU7 😅
Happy to see a @UniFAU alumnus win the best paper award at this year's #IJCNN2023 conference. Also think he'd deserve another award for most hyphens in a paper title 😉 https://twitter.com/RichardGerum/status/1674143313828429826
@GeyserTimes has received an honorable mention in this week's #Yellowstone Caldera Chronicles. An excellent weekly read for anyone interested in the work done by the #YVO by the way. Check out the full article here 👉 https://t.co/REiplfVrHs
Want to score high with your machine learning skills? Then join us for The Tensor Tournament @UniFAU! 3 ML challenges, free food and drinks, and amazing prices await you. Register now for your chance to win Coral AI dev boards. 👉 http://mad.tf.fau.de/t3