P2-0041 Računalniški sistemi, metodologije in inteligentne storitve

Basic information

Abstract

The proposed research would investigate the common characteristics of unstructured and heterogeneous data streams frequently emerging in computer science (e.g. on the World Wide Web, in Earth observation systems, and biomedical systems) that pose immense challenges with their diversities, dynamics, and huge data sizes. With the common goal of unifying their processing at a high level of abstraction, the individual data sources or streams, frequently embedded within strong environmental and instrumental noise, would be decomposed into basic semantic primitives (symbols), denoised and efficiently structured for their enrichment. The supporting algorithms would be implemented as loosely-coupled services, organised into three-tier architecture, and orchestrated for achieving a broad palette of applications from various domains. The first layer would perform domain-specific data management tasks, providing the middleware services from the second layer with interoperable data access. The second layer would be dedicated to the data enrichment and assessment of basic semantic primitives out of the data streams. For this purpose, the primary research focus would be directed towards two recently proposed paradigms: algebraic formalization, of attribute filters based on mathematical morphology, and latent components analysis. The first paradigm would enable precise evaluation of geometric properties by selective and fully automatic adaptation of the investigated patterns to the input datasets, whereas the second paradigm would exploit the time-space dependencies of data symbols for separation of compound data streams into contributions of different sources. Machine learning algorithms would be used to assess the heuristic knowledge about the characteristics of obtained data primitives (symbols) and integrate them within the high-level semantic units. The last layer in this architecture would be the application layer. Here, the middleware services would be chained and integrated into diverse applications, demonstrating universality and functional interoperability of the proposed approaches. As proof of concept, we would implement the detection of irregular muscle contractions from non-invasively acquired surface electromyograms, and monitoring of the Earth’s surface alterations due to landslides or erosions caused by water or wind. Both applications would address the actual socio-economic challenges caused or emphasised by recent demographic and climate changes. The suitable information support, based on elaborated data collection and trusted data interpretation is of key importance for efficient decision-making strategies at national and European levels. The suggested research of the computer algorithms would enable more effective, reliable, and efficient processing of data regarding the addressed applications and would thus significantly contribute to numerous scientific disciplines, also.