Many believe that optimizers like Muon perform well because of their connection to spectral geometry.  But this is not the case! In fact, replacing the spectrum of the update with random or even inverted singular values performs remarkably similar!  arxiv.org/abs/2605.11181

Researchers led by @deltadahl from @CamImaging at @Cambridge_Uni, with @ucl and @QMUL created a system that uses #generativeAI to study the shape and structure of blood cells, and spot unusual or rare cells that may indicate disease such as #leukaemia. dlvr.it/TPSHBw

The Art of Inpainting by Simone Parisotto, Patricia Vitoria, Coloma Ballester, Aurélie Bugeau, Suzanne Reynolds and Carola-Bibiane Schönlieb Explains, reviews and compares different inpainting imaging restoration methods in the challenging scenario... 📚 cup.org/4n9tURh

🎉 Congrats to Yichen Chen — Runner-up for Most Impactful Poster at UoC-AZ PhD Symposium! 🏅 "Automated Discovery of Real & Interpretable Image Biomarkers" And to Simon Deltadahl — CSAR PhD Award 2025 for "Deep Generative Classification of Blood Cell Morphology" 👏

We were delighted to have guest speaker Xue-Cheng Tai from @NORCEresearch present his research on PottsMGNet: A Mathematical Explanation of Encoder-Decoder Based Neural Networks. Fascinating insights into the theoretical foundations of #DeepLearning architectures! 🔍 #AIResearch

🔬Exciting Friday seminar! Fan Zhang presented his production-ready Federated Learning platform for healthcare applications, while Zhongying Deng shared insights on Bilevel Hypergraph Networks for Multi-Modal Alzheimer's Diagnosis. Two fantastic talks advancing #HealthcareAI

If you are coming to #ICLR2025 come say hello. We will be presenting on how to turn Neural Operators equivariant even when all you know is the lie algebra! Generally happy to chat all things #geometry and #optimization in deep learning! 🔗arxiv.org/pdf/2410.02698

🚀 In today's CIA seminar, Leonardo Del Grande (University of Twente) shared valuable insights on Exact Sparse Representation Recovery in convex optimization. A fascinating deep dive into the power of sparsity! 💡🔍 #Optimization #SparseRecovery #MachineLearning

In SIAM News Online, Angelica @aviles_rivero and Carola-Bibiane Schönlieb (@caromitreka) describe semi-supervised hybrid modeling in #healthcare, which preserves theoretical guarantees while achieving higher performance than pure #deeplearning models. siam.org/publications/siam-n…

🩸📝 The BloodCounts! and Flower Labs are delighted to announce the first results of their year-long collaboration today! This collaboration brings together the largest global network of Full Blood Count (FBC) test data and research. Together with Fan Zhang, ML Scientist and Michael Roberts, Chair of BloodCounts! (@blood_counts), we will enable the early and accurate identification of various hematological conditions, such as iron deficiency and anemia, which affect over 1.9 billion people worldwide while still ensuring data privacy and security. 💪🏼 Read the press release for more details! Join us at Flower AI Summit 2025 to learn more about it 🏃🏽‍♀️

In today's seminar, we were thrilled by @JulianMaths's presentation on Validation Data Overuse & Translation into Practice. Insightful discussions and key takeaways on ensuring robust model evaluation! 🚀 #AI #MachineLearning #ModelValidation"

We had an exciting CIA seminar this week! @ChriseldaOliver presented the Linear Wasserstein Framework, and @WallacePeaslee shared fascinating insights on degraded music manuscripts and potential contrast. Great discussions all around! #MusicRestoration #Wasserstein

🚀In today’s CIA seminar, we were delighted to have @AnderBiguri discuss inverse problems, their role in CT/PET imaging, and the latest PET advancements. @KellyKokka also shared insights on using computer vision to extract environmental characteristics from imagery. Great talks!

I recently learned about a very cool fact: if you compose flows of two Hamiltonians, this is also Hamiltonian. What is more - you can write down the exact Hamiltonian!

🚀 In today's CIA seminar, @TheSeparatrix shared insights on detecting iron deficiency in blood donors using full blood count data, while @_JRowbottom introduced G-Adaptivity: an optimised graph-based mesh relocation technique for the finite element method. Fascinating talks!

Fantastic talk from @Zakobian at today's CIA seminar! 🚀 He shared insightful work on the Future of Synthetic Data, covering Model Collapse & Equivariant Neural Operators. 🔥 #AI #MachineLearning #SyntheticData

Dr Anna Breger, working with Professor Carola Schönlieb @caromitreka at the @CamImaging group in @FacultyMaths, will lead a project looking at imaging & ML methods for reconstructing degraded historical sheet music.

Today's CIA seminar was packed with insights! 🎓✨ 🔹 @ChriseldaOliver presented her insightful work on Laplace Learning in Wasserstein Space. 🔹 @WallacePeaslee shared fascinating work on Registering Archival Photos of Panel Paintings. #MachineLearning #ImageProcessing

📣 Registration to the NeurIPS meetup@Cambridge on Fri 6 Dec is now open! If you would like to discuss the ML research being presented at #Neurips2024 register here: buff.ly/3CpXQWw. Further details of the event can be found on the website: buff.ly/3YZFzbj

In today’s CIA seminar, we were thrilled to welcome Christina Runkel for an insightful talk on Reconstructing Training Data🤖. Huge thanks also to Erik Jansson from @goteborgsuni, who shared his research on Geometric Shape Matching for Single-Particle Cryo-EM Data🧬🔬. #AI