Thomas Schnake

We have a new paper on diffusion!📄 Faster diffusion models with total variance/signal-to-noise ratio disentanglement! ⚡️ Our new work shows how to generate stable molecules in sometimes as little 8 steps and match EDM’s image quality with a uniform time grid. 🧵

❓ Do histopathological foundation models eliminate batch effects? ❓ The surprisingly clear answer is: they do not! Find out more in our new paper that we will present at the AIM-FM Workshop at @NeurIPSConf. Link: arxiv.org/abs/2411.05489

Our work on historical insights at scale using machine learning is now out in @ScienceAdvances! Very proud of this team effort, bridging disciplines and institutions—@MPIWG @TUBerlin @bifoldberlin @ml_tuberlin 📜science.org/doi/10.1126/sc…

Big thanks to my collaborators and mentors, first and foremost @IdajiMina, as well as @EberleOliver, @thomschnake, @jdppel, @lciernik, Oliver Buchstab, @am0ck, Frederick Klauschen, Klaus-Robert Müller, and all the supporters from @bifoldberlin and @ml_tuberlin.

xMIL: Insightful Explanations for Multiple Instance Learning in Histopathology by @hense96 @IdajiMina @EberleOliver @thomschnake @jdppel @lciernik , Oliver Buchstab, @am0ck , Frederick Klauschen, Klaus-Robert Müller

✨ Our xMIL paper has been accepted at #NeurIPS2024 ✨ arXiv preprint: arxiv.org/abs/2406.04280

🥳👏Warmest congratulations to BIFOLD scientist Dr. Alexander von Lühmann, who was awarded an ERC Starting Grant of EUR 1.65 million. We are very proud and excited for him and his entire team!!🍾 Read more: bifold.berlin/news-events/ne…

Check out our new position and review paper on representational alignment. We try to find common language across research disciplines, provide a general formal framework, review the existing literature, and discuss open challenges. Amazing team effort!!! 🤖🧠

We are currently preparing the call for papers for Re-Align: The First Workshop on Representational Alignment at #ICLR2024! You can shoot me a DM (or fill out the form below👇) if you want to be on the reviewing committee.

We're excited to present our paper "Fast and Unified Path Gradient Estimators for Normalizing Flows" at #ICLR2024 in two weeks! 🎉 This work focuses on deriving efficient gradient estimators for coupling-type normalizing flows using the inverse function theorem.

Faithful attributions are measured by pixel flipping! But how trustworthy and consistent are the PF benchmarks? Read our new preprint to learn more! @nstrodt @jvielhaben @ml_tuberlin @bifoldberlin @FraunhoferHHI #XAI #diffusion arxiv.org/abs/2401.06654

@haokeqiu @gklambauer Thanks for your interest @haokeqiu. Yes, this data is in principle usable. There is no restriction on any data type. We didn’t test it on house price-like data, though.

@gklambauer Excellent work! Can it be used in tabulate data, such as a Boston House Price-like dataset?

Towards Symbolic XAI -- Explanation Through Human Understandable Logical Relationships Between Features Abstractions on top of traditional XAI methods are used. These are combined with logical operators to provide explanations... P: arxiv.org/abs/2408.17198

Hey @gklambauer thanks for reposting. For me as an author the preparation was a great experience! If you have any questions #symbolicXAI #symbXAI feel free to ask me directly.

🎉Our preprint <<Towards Symbolic XAI>> came out! Next to classic ML stuff like NLP, we have a chemistry section! ⚗️ What's going on inside a neural network while it's predicting molecular properties? arxiv.org/abs/2408.17198

📢 New Preprint Alert 🚨 Ever Wondered How to Make Mamba Models More Transparent? Meet MambaLRP🔥 We introduce MambaLRP, a novel algorithm within the Layer-wise Relevance Propagation (LRP) framework, specially designed to enhance the explainability of Mamba models. 1/7

Enjoying my visit at Chemical Compound Space Conference (CCSC) in Heidelberg. #chemistry #machinelearning

Mhh... where does that come from? heise.de/news/Machtkamp…

We have a new website 🎉 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