I build the self-explaining hospital - interpretable, privacy-preserving clinical AI.
Block-term and multiway tensor models whose decomposition is the explanation - structure a clinician already reasons about, read straight off the model rather than approximated afterward. The methodological core under everything else.
Training shared models across institutions without raw data ever leaving them - including the hard case of different modalities and incomplete records, with privacy built in by design.
Backbones that learn from unlabelled, longitudinal data and fuse imaging, omics and clinical signals into a single shared representation for downstream clinical prediction.
Recovering interpretable spatiotemporal structure from high-density brain signals (ECoG / EEG) with tensor regression - where the methodology was forged before it moved to the clinic.
I am open to collaboration on interpretable and federated clinical AI, and to supervising motivated students. The fastest way to reach me is email.
axel.faes@utwente.nl