The research topic “Representation and data mining” is involved in the design, development and assessment of semi-supervised tools for the description, segmentation, classification, analysis and visualisation of data adapted to extremely large multimedia databases with a view to facilitating their access, consultation and control.

The originality of this research topic lies in the fact that it addresses two distinctive features of the Ina context. INA has 150 expert researchers. Our goal is to assist those experts by providing HCI based on interactive visualisation and apprenticeship systems. That approach is reinforced by a policy ensuring the thorough evaluation of prototypes developed for those experts. The collections of INA and the Inathèque are made up of hundreds of thousands of hours of material that has already been archived and about 100 incoming channel feeds. Taking into account the scale of data to be dealt with is therefore at the core of this work.

The research activities cover the design of multimodal descriptors and similarity measures, research technologies for extremely large databases of descriptors, the design of active learning methods, visualisation and HCI. The modules developed are integrated into our internal experimental prototypes and/or national or European projects and allow us to create various applications such as the analysis of newsreel databases, the structuring of television feeds, programmes or collections, the classification of images, or multimodal research engines.