Project: EDL P16-25-P4
Project title: Deep Learning for High-Tech Systems and Materials (HTSM)
Project leader: Prof. Dr. Rob van Nieuwpoort (UvA, NLeSC)
Co-applicant(s): Prof. Max Welling (UvA), Prof. Henri Bal (VU), Prof. Henk Corporaal (TU/e)
Partners: FEI, Scyfer, Tata Steel, NLeSC, Astron, SURFsara
Short description: DL for large scale fabrication processes and large-scale scientific instruments (steel inspection, surveillance, electron microscopes, telescopes, detectors)
P4 addresses four main data efficient DL approaches:

  • Deal efficiently with combinations of large, sparse, heterogeneous, and multimodal data.
  • Combining semi-supervised DL techniques with active learning to reduce the amount of labeled data needed, and deal with extremely unbalanced clusters.
  • Combine deep generative nets and ladder nets with active learning to reduce the amount of labeled data needed, and deal with extremely unbalanced clusters.
  • Exploit accelerators, special-purpose hardware, and distributed services to enable real-time active learning.