I am a Ph.D. student at the machine learning group at the Institute of computational biology, Helmholtz Center Munich. My interests are developing machine learn
Our innovation grant-winning study on AI-based OCT analysis post-PCI is on medRxiv! DeepNeo is a tool for neointima & stent analysis with a GUI (⬇️) ! It classifies neoint. w same acc as experts + strong segmentation of lumen, stent and neoint.🔬 1/6
https://t.co/AfjwaG3RC9
Excited that our paper on data-efficient self-supervised learning is out @NatMachIntell. Led by @holmberg_olle, we apply this to Diabetic Retinopathy screening, needing 75% less annotations & 200x fewer params to reach human performance than before.
https://t.co/UIy33Htkz0
Want to map #scRNA-seq data on a reference atlas without sharing raw data? Plz check out single-cell architectural surgery (scArches) - led by @MohammadLotfol1, we use transfer learning for efficient query to reference integration. https://t.co/pvl6alBXoI https://t.co/HEO2RPikJT
Really happy that our paper on the proteome landscape across 100 species is out @Nature. Led by @labs_mann, we contributed a machine-learning based exploration of that landscape using a bidirectional LSTM, and robustly predicted a peptide's retention time. https://t.co/zFNRJJgY3e
Excited to share our work on benchmarking large-scale data integration in single-cell genomics. Led by @MDLuecken and starting from an internal hackathon with @colomemaria_epi, we compare methods & preprocessing using a series of metrics and scenarios, see https://t.co/wAxIq0GTRm
Here's one of my favorite poems by Charles Bukowski, titled Nirvana, about the fleeting moments of magic in life. I've written many songs, most never shared. I hope to change that. Me playing covers & reading this poem is a step toward it. Happy holidays. https://t.co/IYw0ohfzB2
Unsupervised pre-training now outperforms supervised learning on ImageNet for any data regime (see figure) and also for transfer learning to Pascal VOC object detection
https://t.co/MvRu3dqTxk
How to deal with costly annotation to apply #DeepLearning to #medicalimaging? In a work led by @holmberg_olle and Niklas Köhler, we leverage self-supervised learning to predict retinal depth from fundus alone, together with @LMU_Muenchen eye clinic. https://t.co/AJct0AdnNK
Excited to share our scVelo manuscript - led by @VolkerBergen and @falexwolf, we generalize the beautiful RNA velocity concept from @KharchenkoLab and @slinnarsson to transient cell states through dynamical modeling. https://t.co/Dv11ChjUGW and https://t.co/Dl6oR41VxG #scRNAseq
What makes a great paper? Julia Eckhoft shares her advice on writing a great research paper to please readers, editors and yourself.
https://t.co/V57H3Y1FJt
@NatureComms
Open discussion of differing world views is essential for progress. Always eager to have my ideas challenged.
I invite @geoffreyhinton to help raise level of discussion via public conversation @letterwiki
RT if you would like to see this happen https://t.co/S3EEKffAup
Excited to share our recent work on generative modeling for unpaired data using a ‚transformer VAE’. Led by @MohammadLotfol1 and @falexwolf, we extend scGen using a conditional VAE together with an MMD regularization. Applications for images and scRNA-seq. https://t.co/YNqL37lZqV