Research
Scalable generative models for high-dimensional, low-sample-size domains.
Danilo DanesePolitecnico di Bari · SisInfLab
Previously: Visiting researcher, University College London; Queen Square Analytics
Research on generative models and representation learning in high-dimensional, low-sample-size (HDLSS) domains, with 3D neuroimaging as the main setting: how to generate realistic volumetric data from few subjects, how to do so within modest compute, and how to verify that what a model generates or explains is reliable and not merely plausible.
The same questions extend to explainable AI for clinical prediction and to multimodal recommender systems, where supervision is sparse. Academic research is complemented by applied projects with industry partners and by the operation of multi-GPU and HPC infrastructure for large-scale model training.
Flow matchingDiffusionWavelets3D brain MRIBrain-age modelingExplainable AIMultimodal recommendation