🧵I am excited to share that our paper: AI-Based Anomaly Detection for Clinical-Grade Histopathological Diagnostics has now been published in @NEJM_AI .
ai.nejm.org/doi/full/10.10…
ai.nejm.org/doi/full/10.10…
Jonas Dippel
@jdppel
PhD Student at @ml_tuberlin, @bifoldberlin and Data Scientist at @aignostics | Training large neural nets for computational pathology.
416 Following 108 Followers 57 Posts
Latest posts 5
We built a new website for our lab🧑💻 Check out researchers, publications, courses and software from our group!
We have a new website 🎉View quoted post on X
Thanks to all the people who put countless hours into making it look as amazing as it does!
Want to learn more about our team, #research, open positions and offered courses?
Check it out 👉 web.ml.tu-berlin.de
#MachineLearning #Berlin
I will present a demo showcasing #XAI for histopathology tomorrow in the @TUBerlin main building.
#LNDW23
#LNDW23
BIFOLD will exhibit 2 showcases during @LNDWberlin: 1⃣CORALS and
2⃣AI Made in Berlin, which features groundbreaking research in #cancer, earth observation (#eo) and #IoT.
Join us and explore. t1p.de/jnrj4
📆Sat, June 17
⏰5 - 10pm
🏛️@TUberlin main building
#LNDW23View quoted post on X
We learn a linear transformation that aligns image-text models with human similarity judgements. By keeping the local structure intact while improving the global structure of the representation space, we improve few-shot and AD downstream task performance.
🚨Beep beep 🚨 Have you ever been wondering about how to use human similarity judgments for improving neural network representations? We have something for you! We found a transform that improves representational alignment and downstream task performance!View quoted post on X
arxiv.org/abs/2306.04507
Our paper for #ICLR2023 which studies human alignment of vision models is now updated with interesting additional experiments.
🥳I am delighted to share the camera-ready version of this work for ICLR 2023. It has been an intense rebuttal but due to additional experiments that we were required to do, we’ve actually learned a few new interesting things that we did not know before.View quoted post on X
arxiv.org/pdf/2211.01201…