Section: Other Grants and Activities
Collaboration with Reykjavík University, Iceland
Participant : Laurent Amsaleg.
This collaboration is done in the context of the INRIA Associate Teams program. This program links two research teams (one INRIA, one foreign) willing to cross-leverage their respective excellence and their complementarity. Björn Þór Jónsson (Associate Professor) leads the team of researchers involved in Iceland.
Image databases, and content-based image retrieval systems in particular, have become increasingly important in many applications areas. While extremely effective (they return high quality results), these systems are very inefficient (they answer very slowly) due to their complexity, to the curse of dimensionality problems and to the scale at which they have to run when dealing with collections of realistic sizes.
Originally, the goal of this project was to research and develop new database support that integrates efficiency and effectiveness for modern, large-scale, computer-vision related applications and problems. We found, however, that the inefficiency observation applies not only to images, but to most multimedia documents as soon as the retrieval is based on the similarity of low-level descriptions.
Therefore, the Eff2 associate team has widen its scope and tackles efficient database support for more general large-scale multimedia applications and problems that mostly deal with high-dimensional low-level features. We now investigate additional issues such as efficient query execution on large collections of sequences of low-level descriptions (e.g., audio or video collections). We have also initiated another thread of research by investigating the browsing of personal image collections. Going to sequences and addressing browsing issues (and not solely searching) is a new development in our cooperation.
Collaboration with Croatia and Slovenia
Participant : Annie Morin.
Medical School, University of Zagreb, department of Electronics, Microelectronics, Computer and Intelligent systems, University of Zagreb, Zagreb, Croatia; Faculty of Computer and Information Science, University of Ljubljana, Slovenia; ERIC lab., University of Lyon2
A. Morin got two Egide contracts with Slovenia (Proteus) and Croatia (Cogito) for 2007 and 2008 on knowledge discovery and visualization for textual data. In Slovenia, we work with Blaz Zupan and Janez Demsar from faculty of Computer and Information Science, University of Ljubljana and in Croatia with Bojana Dalbelo Bašić from faculty of Electrical Engineering and Computing, university of Zagreb. The French laboratory ERIC (university of Lyon 2) is the other French partner.
The concerned research teams have different expertise on the same subject: machine learning for the Slovenian and Croatian teams, statistics for the French teams and common abilities such as development of open source data mining software and visualization tools. They have been in touch since a first meeting in 2004 on intelligent data mining. We plan to implement a new visualization system for textual data. Proposed collaboration includes sharing of a number of Ph.D. students.
In spite of the end of our Egide collaboration, two papers were presented this year, the first one with Slovenian team, dealing with subgroups discovery when there exists dependencies between explanatory variables  . The second one deals with textual data analysis and correspondence analysis. It investigates how the document representations (words, word n-grams, letter n-grams) influences the quality of visualization  . Additionally, Artur Šilić from the university of Zagreb continued his work on temporal changes in text streams using the New York Times corpus.
Collaboration with Nagoya University and National Institute of Informatics, Japan
Participants : Laurent Amsaleg, Patrick Gros, Fabienne Moreau, Annie Morin, Pascale Sébillot, Julien Fayolle.
In the framework of our collaboration with both the University of Nagoya (Ichiro Ide, Associate Professor) and the National Institute of Informatics in Tokyo (Shin'ichi Satoh, Professor), Julien Fayolle spent 4.5 months in Japan for a joint Master's degree internship dedicated to topic segmentation of TV broadcast news, integrating video clues, and to topic-threading.