Horizons for Information Societies
Seminar #1

The 1st seminar on Horizons for Information Societies was about data processing. Dr Grira () and Dr Martinet (, ) presented their research on, respectively, semi-supervised clustering and inter-modal analysis at the .

See also: next seminar (#2).



Slides of the presentations:

Date: 1 November 2006 (15:00-17:00)
Location: , Tokyo, Japan
Language: English
Registration fees: None
Organization: Dr ()


Semi-Supervised Clustering in Image Databases
by Dr

Abstract: The general purpose of clustering is to organize data into groups, in an unsupervised manner such that items within a group are more "similar" to each other than to items from other groups. This talk will consider the situation where some a priori in the form of constraints are available to the clustering algorithm, and explore a method taking into account this additional information. The resulting combination of this semantic-based and feature-based sources of information form a growing field called semi-supervised clustering.

Speaker: Project Researcher at the , Dr Grira obtained his Ph.D. in computer science from university (France). His Ph.D. work was achieved between 2003-2006 within the IMEDIA research group at (the French National Institute of Computer Science and Control) and his thesis title was: Semi-supervised Clustering in Large image databases.


Inter-Modal Analysis for Document Annotation: an Information Theoretic Approach
by Dr

Abstract: Inter-modal analysis consists in modelling and learning relationships between several modalities from a data set in order to extract semantics. I will introduce the concept of inter-modal analysis for the task of document annotation, in the context of visual and textual modalities, using extracted semantics to predict a textual annotation given a piece of visual data. I will describe an information theoretic approach based on two sources: (1) the mutual information of modalities and (2) the information entropy of the distribution of the visual modality against the textual modality. I will show how such an approach can efficiently annotate both image and video documents.

Speaker: Dr Martinet is a , working at on the processing of multimedia data.