iiinova — where your digital self keeps evolving

Jakob Ambsdorf

@jakamb.bsky.social

40 followers 87 following 0 posts

PhD Fellow @ DIKU/Pioneer Centre for AI scholar.google.com/citations?user=C…

Posts Replies Media

Reposted by Jakob Ambsdorf

Stefano Cerri @stefanocerri.bsky.social · 21/01/2026

Introducing FOMO300K — a large-scale, heterogeneous brain MRI dataset for self-supervised learning 🧠 318,877 scans from 920 datasets, accessible in a single click via 🤗 Hugging Face 📄 Preprint: arxiv.org/pdf/2506.14432 📊 Dataset: huggingface.co/datasets/FOM... 💻 Code: github.com/Sllambias/as...

1133

Reposted by Jakob Ambsdorf

FOMO25: Foundation Model Challenge for Brain MRI @fomochallenge.bsky.social · 19/08/2025

Deadline extended! Interested in foundation models for brain MRI? You still have time to participate in the FOMO25 Challenge! Deadline has been extended until August 31, 2025. To get started, check out: 🌍 website: fomo25.github.io 💻 code: lnkd.in/d8W8k9aV 🤗 checkpoints: lnkd.in/dZESityQ

fomo25.github.io

FOMO 2025 - Foundation Model Challenge for Brain MRI

023

Reposted by Jakob Ambsdorf

Pioneer Centre for AI @aicentre.dk · 17/07/2025

Training a custom foundation model for your niche imaging domain? In our #MICCAI2025 paper, we demonstrate on fetal ultrasound: A smaller model on less data wins when trained in-domain, while generalist models fail to show benefits of scaling. arxiv.org/abs/2506.19552

arxiv.org

General Methods Make Great Domain-specific Foundation Models: A Case-study on Fetal Ultrasound

With access to large-scale, unlabeled medical datasets, researchers are confronted with two questions: Should they attempt to pretrain a custom foundation model on this medical data, or use transfer-l...

172

Reposted by Jakob Ambsdorf

FOMO25: Foundation Model Challenge for Brain MRI @fomochallenge.bsky.social · 22/04/2025

FOMO25 accepted at #MICCAI2025! 🚀 Excited to announce the first challenge at MICCAI focusing on the development of self-supervised pretraining of foundation models for brain MRI! 🧠 With access to a large-scale dataset, codebase, cash prices and multiple tracks. Read more ⬇️

1118