Vous êtes ici

Séminaire du jeudi 20/11/2014

Date: 
Jeudi, Novembre 20, 2014 - 14:00 to 16:00
localisation: 
Amhi IMB, Bat A33

Orateur : Alan Hanjalic

Affiliation : TU Delft, Netherlands

Titre : On Intent-Aware Video Search

Résumé :
In this talk I will address the research that aims at advancing video search engines by making them aware of user intent i.e., the immediate goal a user seeks to accomplish with video search. Making video search engines intent-aware is a challenge because user intent is often not explicitly reflected in the query. While the query typically reveals WHAT the user is searching for (the “topical” dimension), the intent reflects WHY she has inserted that query in the first place. Since a user’s information need is determined by both the WHAT and WHY dimension, a search engine can fulfill this need to its full extent only if the relevance models take both dimensions into account. Since the effort invested in the development of video search engines has focused mainly on the WHAT dimension, multimedia research has yet to turn its attention to developing algorithms sensitive to why users are searching for video in order to provide more articulate search results. In this talk I will present first findings on how to enable this intent-related focus shift in the design and realization of video search engines.

A propos de l'orateur :
Alan Hanjalic is a professor of computer science and head of the Multimedia Computing Group at the Delft University of Technology, The Netherlands. His research focus is on multimedia search and recommendation. Prof. Hanjalic has been a member of Editorial Boards of several scientific journals in the multimedia field, including the IEEE Transactions on Multimedia, the IEEE Transactions on Affective Computing and the International Journal of Multimedia Information Retrieval. He was also a member of the IEEE Technical Committee on Multimedia Signal Processing. As of 2014, he is Chair of the Steering Committee of the IEEE Transactions on Multimedia and Associate Editor-in-Chief of the IEEE MultiMedia Magazine. He was a keynote speaker at a.o. the IEEE International Workshop on Multimedia Signal Processing 2013, International Conference on Multimedia Modeling 2012, and the Pacific-Rim Conference on Multimedia 2007. Prof. Hanjalic has held key positions in the organizing committees of leading conferences in the multimedia field, among which the ACM Multimedia (General Chair 2009, Program Chair 2007 and 2014), ACM ICMR (Program Chair 2011), ACM CIVR (Program Chair 2008) and IEEE ICME (Program Chair 2012). Prof. Hanjalic will be a General Chair of the ACM Multimedia conference, Amsterdam, The Netherlands, 2016.

--------------------------------------------------------------------------------
Orateur : Luc Brun

Affiliation : Ensicaen

Titre : Graph kernels in Chemoinformatics.

Résumé :
Chemoinformatics is the use of computer and informational techniques applied to a range of problems in the field of chemistry (Wikipedia). This broad subject includes various topics such as the drawing of molecules or the creation of databases of molecules accessible through web search engines. In this talk we will explore a much more restricted field devoted to the prediction of molecule's properties through machine learning techniques. One of the main principle of chemoinformatics states that "two similar molecules should have similar properties". The methods presented in this talk will thus be based on the design of similarity measures between molecules. Such similarity measures are encoded by graph kernels based on two important properties of molecules: 1) The types of their atoms and the type of bounds between these atoms and 2) the different cycles of molecules and their adjacency relationships. The use of graph kernels allows to combine graph similarity measures with usual machine learning methods. We will present some experiments and conclude this talk on some extensions of this work to other fields.