Pioneering data science and computation approaches
IIS's aim at resolving several grand challenges for sustaining civilization’s continuing advancement while still improving quality of life as a whole. The posed challenges, such as reverse engineering the brain, advanced personalized learning, and engineering the tools of scientific discovery, make it clear that sensors, signal processing hardware, and mathematical algorithms are now under increasing pressure to accommodate higher dimensional data sets; ever faster capture, sampling and processing rates; ever lower power consumption; communication over ever more difficult channels; and radically new sensing modalities. However, the fundamental issues that trigger IISs have not been examined in a collective and coherent way by the scientific community. To advance IISs, various stakeholders must work together to establish a framework that supports efficient sensing, processing, data fusion, decision making, direct performance analysis and prediction. With this in mind, the EU-funded SUBSPARSE (Geometric and combinatorial foundations for emerging information and inference systems) project set out to devise a theory that delivers positive outcomes for different IIS issues by effectively using existing resource networks. Project partners created innovative theories and algorithms by combining concepts from statistical signal processing, geometrical modelling and combinatorial optimisation. Their approaches were implemented in an overall platform from which to deal with several signal processing and machine learning problems found in different data modalities. Key outcomes include the first-ever learning framework to create signal dictionaries for sparse representation and a mathematical framework that obtains sparse projections onto convex sets. The team characterised structured sparse sets and established a mathematical link between compressive and game theory. Nearly 100 papers were published in various leading journals and presented at high-profile international conferences and workshops. The frameworks and algorithms developed in SUBSPARSE will set in motion novel approaches to such fields as machine learning, robotics, virtual reality, remote sensing and bio-imaging.
Keywords
Data science, computation, information and inference systems, geometrical modeling, combinatorial optimisation
