The EDL program is set up as a matrix of science-driven research lines (RLs) and application-inspired projects. These RLs are needed to ensure and stimulate synergy between the projects. All projects focus on making Deep Learning efficient, attacking all relevant levels, and demonstrate quantifiable efficiency improvements in their use cases. The involvement of each project within the RLs is indicated (in PhD/Postdoc/PDEng FTEs) in the matrix. Together the EDL projects will show that DL can be easily and efficiently used in many application domains, especially the ones important for Dutch economy and industry. In particular:
P1. Demonstrates fast realization of DL use cases by High Performance Computer services.
P2. Demonstrates DL for improved 3D modeling and reconstruction.
P3. Improves functionality of video analyses for camera surveillance using DL.
P4. Develops DL techniques for product inspection in next generation high-tech systems
P5. Realizes fully automatic health monitoring for humans and animals.
P6. Designs more reliable vehicle robotics used in unstructured environments.
P7. Designs energy-efficient processing HW/SW platforms and mapping methods for DL algorithms.
Projects P1 and P7 serve a program wide role. P1 develops a Dutch infrastructure providing DL as a service to make EDL technology easily available to a range of users and enterprises; the Dutch SURFsara and eScience center are instrumental in making these services available. P7 implements DL methods, developed within other projects, in energy efficient ways on advanced and new processing platforms.
P# | Name + Project Leader | Short description | Use cases |
P1 | DL as a Service (DLaaS) PL: Mark Hoogendoorn (VU) |
Tooling to improve the usability and efficiency of deep learning for third-parties | Financial data analysis for a bank
Predictive maintenance for electromotors |
P2 | DL for 3D reconstruction, matching and recognition (RMR) PL: Theo Gevers (UvA) |
Learning methods for reconstructing 3D computer graphics models directly from sensory data | 3D Interior reconstruction,
3D cadaster creation and tooling |
P3 | DL for Video Analysis and Surveillance (VAS) PL: Gijs Dubbleman (TUe) |
DL for automated video & streaming data analysis | Airport surveillance Fully automated driving |
P4 | DL for High-tech Systems & Materials (HTSM) PL: Rob van Nieuwpoort (UvA) |
DL for large scale fabrication processes and large-scale scientific instruments (steel inspection, surveillance, electron microscopes, telescopes, detectors). | Cryo-electron microscopy
Steel surface inspection Radio astronomy image denoising |
P5 | DL for Human and Animal Health (HAAH) PL: Henkjan Huisman (Radboud) |
DL for automatic monitoring and diagnostics for medical diagnostics | Health monitoring of humans and
Health monitoring of groups of cows |
P6 | DL for Mobile Robotics (MobRob) PL: Dariu Gavrila (TUD) |
Learning methods to allow mobile robots to perceive and act intelligent on their environment | Mobile robotics |
P7 | DL HW-SW platforms (DLPL) PL: Sander Stuijk (TUe) |
HW-SW design for efficient processing of DNNs low energy embedded systems high performance | Small mobile platform for Deep Neural Network applications |