Markus Frey

Our paper "Magnetic resonance-based eye tracking using deep neural networks" is now published in @NatureNeuro Paper: Code: Source Data & Model weights:

#DeepMReye is out! Use #deeplearning to perform #eyetracking in #fMRI without camera! @CYHSM & I are thrilled to finally share our Code + Data + Notebooks + User Documentation + Paper @NatureNeuro! With @doellerlab @KISNeuro @MPI_CBS Thread below!👇1/7

RT @effi288: 📄 Releasing the Soofi S pretraining tech report: a sovereign, open foundation model for German and English Today we’re publishing the full pretraining tech report and project page for Soofi S 30B-A3B — a Mixture-of-Experts hybrid Mamba model trained on ~27 trillion tokens with deliberately up-weighted German. What’s in the report: 🏆 Strongest fully open model in our evaluations on BOTH the English and German aggregates — ahead of Olmo 3 32B and Apertus 70B (full methodology in the report) 📋 Radical transparency: complete per-source data accounting, all hyperparameters, training + eval code, checkpoints — everything under permissive licenses 🇩🇪 Trained end-to-end on Deutsche Telekom’s Industrial AI Cloud in Munich — sovereign AI infrastructure on German soil Soofi S combines frontier-level capability with the highest measured aggregate long-context decode TPS, and unlike full-attention dense baselines maintains high throughput as context grows. The figure plots Capability Index versus measured aggregate decode TPS/GPU at 40K context and batch 32. The Capability Index averages five benchmark groups, i.e., Code, GSM8K, GPQA-Diamond, English aggregate, and German aggregate, after normalizing each group to the best plotted model. Aggregate decode TPS/GPU is measured with a TP=1, one-B200 vLLM latency-subtraction protocol.

I will be presenting this at the ICLR 2026 Latent & Implicit Thinking Workshop. Paper Link: We are also hiring at Lamarr Institute. If you're interested in working on these kinds of problems, feel free to reach out.

RT @NauMatt: #CogNeuro seeks unified theories of behavioral, physiological, and mental states. To this aim, our @NatureNeuro Perspective proposes a new framework centered on #TaskDemands & across-task generalization. w/@Alex_C_Schmid @smkaps @Chris_I_Baker D.Kravitz🧵

RT @MPI_CBS: Our Otto Hahn Medal 🏅winners @RachelGZsido and Markus Frey @CYHSM explain in a short interview what motivates them in their #research, what their #thesis was all about and what #sciencetopics they are working on atm #sciencegoals:

RT @maxplanckpress: Congratulations to all our amazingly talented young researchers who were honoured with an Otto-Hahn-Medal today for the…

RT @TrackingActions: 🎉 new preprint posted of potential interest, lead by talented masters student @AChRD4: 🔥fully unsupervised 3D cell segmentation that is as good (or better) than leading supervised methods, with @napari_imaging integration! 👏 ➕new 3D GT dataset ⬇️🧵

Have you tried #CellSeg3D yet for automating your 3D lightsheet microscopy analysis? We just hit >75K pip installations! 🥳 We have some 🔥 coming soon ... but in the meantime, lead by @AChRD4 @Neuro_X_EPFL

RT @epstein_lab: The Epstein Lab @Penn is looking for new postdocs!!! NIH funded project to investigate spatial and nonspatial cognitive maps in humans. Collaboration with @NoraNewcombe. More info at the link (please RT):

RT @EdvardMoser: Hiring 6 #postdocs & PhDs to live out your scientific dreams! Theory-guided search for cortical algorithms w/ latest generation #Neuropixels, @TheMini2P & advanced data analysis at @KISNeuro, with #KiloNeurons, @MayBrittMoser & @WikiZong Come join us!

RT @AndrejBicanski: Dear colleagues, please circulate these ads for two fully funded PhD positions in my new group. Topics can cover grid cells, retrosplenial cortex, head direction, subiculum and boundary coding + cog maps generally.

RT @TrackingActions: 🚨job alert! Myself, Silvia Arber and Alex Mathis are looking for a postdoc to work with us to do hierarchical modeling of sensorimotor control, using a lot of incredible data 🔥🐭 This builds on our lines of forelimb motor control tasks & modeling! Ad:

RT @TrackingActions: 🔮Ever wished you could just say what behavioral analysis you wanted? ... "Plot the animal's trajectory" "Count the head hips in the open arm" 👀🪄we present AmadeusGPT🎻 to do this ... 🗞@shaokaiyeah @jessy_lauer @zhoumu53 @trackingplumes and me 🧵⬇️

RT @mona_garvert: I am hiring a PhD student in cognitive computational #neuroscience! 👩🏻‍🎓👨🏻‍🎓Come and help us discover how the human brain learns, decides and guides flexible behaviour in beautiful Würzburg 🧠🙌 Find out more & apply here: @bioDGPs_DGPA @FENSorg @SfNtweets

RT @martin_hebart: I am looking for two postdocs to join us at Giessen University @jlugiessen, in collab. with the Max Planck Institute @MPI_CBS in Leipzig, funded by @ERC_Research & @ProLOEWE. We have exciting projects lined up but also offer a lot of freedom with implementing own ideas!👇 pls RT!

RT @AuschwitzMuseum: @elonmusk Support memory by excluding tweets of @AuschwitzMuseum from the limits. Most of our tweets commemorate individual victims of the Auschwitz camp on their birthdays. We remind their names, their faces and their fate. They all deserve to be seen and remembered.

RT @NauMatt: #DeepMReye has a new @streamlit app! 🎊 Camera-less #eyetracking in #fMRI has never been easier. ◦ Install & open ◦ Pick a pretrained model ◦ Upload fMRI data ◦ Download gaze coordinates Kudos to @CYHSM for the app & thanks to all contributors! 1/3

@NauMatt and I are excited to introduce DeepMReye v0.2 🥳Now featuring a local Streamlit app for efficient extraction of gaze coordinates using pre-trained model weights: Please let us know if you run into any problems with this!

A sincere thank you to our open-source contributors: @yarikoptic, @WeeXee and especially @RemiGau for making it BIDS compatible.

RT @CYHSM: Excited to share our work on scene perception in artificial neural networks #CVPR2023. Will talk about egocentric to allocentric reference frame transformations & unsupervised scene segmentation (10.30am - TUE-AM, 202) Can't make it? See Video below👇

Work done with amazing PIs @caswellcaswell and @doellerlab building on top of 4MT from @tom_hartley, @NeilBurgess10, CorNet from @JamesJDiCarlo, @dyamins @martin_schrimpf and unsupervised segmentation models from @tkipf @RishabhKabra @taiyasaki + many others not on Twitter

Original Twitter thread here:

Excited to share the last journey of my Ph.D., accepted at #CVPR2023!🥳 We propose the Allocentric Scene Perception benchmark (ASP) and set new state-of-the-art results on unsupervised scene segmentation tasks. Paper: Thread and Code below 👇