DEDALE

Data Learning on Manifolds and Future Challenges, 09/2015 – 08/2018.

Funded by: European Commission, H2020-FET-OPEN-2014.

Partners: CosmoStat Lab – Service d’Astrophysique – Irfu – CEA, FORTH-ICS (Coordinator,  P. Tsakalides), University College London: Astrophysics Department (UK), Technical University Berlin (Germany), Sagem Défense Sécurité (SAGEM – France).

Funding: € 2,702,398, (FORTH-ICS: €560,000).

SPL Contact Person: Prof Panagiotis Tsakalides

Summary: Modern imaging instruments have gone a long away from their early predecessors in terms of complexity, resolution, accuracy, speed and scale. However, revealing and understanding the acquired scientific signals, comes at a great cost in terms of data analysis requirements. The quality of instrumentation coupled with the data deluge are key drivers in the cutting edge research area of Scientific Big Data, which represents a radical turn in how modern systematic research is conducted.

The key objective of DEDALE is to depart from state-of-the-art sparse linear models to new more physically relevant, non-linear, low-dimensional models for manifold-valued data for analyzing Scientific Big Data. Furthermore, DEDALE will investigate and develop high-performance computing methods for signal learning and data mining tasks such as dimensionality reduction, classification and clustering of such data. Methods developed in DEDALE will be employed for analyzing data from the Euclid space mission in an effort to produce an accurate map of the dark matter distribution in the universe.

The role of SPL:

  • Investigate cutting edge non-linear learning on complex imaging data.

  • Design dedicated solvers for the recovery of multivariate complex signals.

  • Develop large scale machine learning schemes for image and signal processing.

Related Publications:

Karalas, Konstantinos; Tsagkatakis, Grigorios; Zervakis, Michalis; Tsakalides, Panagiotis, Deep learning for multi-label land cover classification, SPIE Remote Sensing, September 2015, Toulouse.

Tsagkatakis, Grigorios; Tsakalides, Panagiotis, Recovery of quantized compressed sensing measurements, IS&T/SPIE Electronic Imaging, February 2015, San Francisco.

Fotiadou, Konstantina; Tsagkatakis, Grigorios; Tsakalides, Panagiotis, Low Light Image Enhancement via Sparse Representations, Image Analysis and Recognition, 84-93, 2015, Porto.

Achievements:

  • The synthesis of the Coupled Sparse Dictionary Learning technique for solving the problem of spectral super resolution.
  • Design and development of the DEDALE distributed learning platform for Big Imaging Data;