Fermer
_fse_imi_choix.jpg

Research

The research activity is mainly focusing on two domains, namely knowledge representation and data mining.
  • In the domain of knowledge representation, a formal work on the trade-off between scalability and expressivity led to a first prototype implementation of a scalable knowledge representation system (PARKA), which was successfully used among others in medical information systems. Over the last years the research activity was oriented toward the possibility of using an ontological approach for modelling multi-dimensional data. Dynamic ontologies refer to the specification of a mechanism for managing reference ontologies and of a formal framework for learning ontology macros and ontology optimization. The projects in this domain conducted to the developement of KnOWLer, an ontology-based information management system targeting semantic integration into large-scale information systems, and of IKARO, a general-purpose formal system for repurposing ontologies. 
  • In the domain of data mining, the research activity primarily focused on two questions: firstly, how data mining algorithms can benefit from knowledge representation systems, and secondly, how the efficiency of existing systems can be improved. A first important contribution in this work consisted in the formalization and the implementation of a knowledge representation language that was scalable enough to be used in conjunction with data mining tools. A very efficient system based on a relational database system and a sophisticated query language was designed and implemented. The principal data mining tasks performed by the system were high level classification rule induction, indexing and grouping. A second contribution is a deep analyse of the criterion used during the construction of a decision tree, which conducted to a new family of split functions with better performance on large data sets than classical split functions.  And a third contribution consists in the formalization of a general framework for temporal data mining which also capture uncertainty aspects of knowledge.

The institute developed and strengthened a policy of collaboration with various institutions, either of type academic, economic or public administration. IMI participated in different CTI research projects of type public-private and initiated collaborative research projects with socio-economic partners: with Institute of Forensic Sciences Lausanne on domain-driven data mining, with Enterprise Institute Neuchâtel on qualitative data management, and with Karman Centre Bern on knowledge management in computer vision. The projects are largely funded by public entities (Swiss National Scientific Foundation, Commission for Technology and Innovation), as well as by private funds.