XRCE organises public scientific seminars on a regular basis which you are welcome to attend. These seminars are an occasion to exchange with researchers from various backgrounds and to broaden scientific expertise. You can subscribe to our seminar RSS Feed for dates, speakers and topics.
Up to now, digitisation is usually understood as the solution for long term preservation: being able to copy bitwise assets, so that new versions are identical to old ones, suggests that we can cope with access (ubiquity of content) and with preservation (replication before medium decay). However, we know nowadays that digitization is the problem. If you want to make a content unaccessible in the future : just digitize it !
Our digital environment entails that we should envisage a new way of preserving. In this talk, after a survey of the main difficulties, strategies and solutions, I will present a patrimonial approach for preservation, where contents can be preserved thanks to a continual access and a cultural activity. Moreover, due to technical complexity, contents should be reinvented to be accessed : then integrity and authenticity should be redefined where new versions created for contemporary access are criticized and annotated according to philological and hermeneutical principles. Preserving is from now on a continuous work on variants.Slides (2.95 MB)
Autonomic computing is a promising solution to ever-increasing system complexity and to the costs of human system management as systems scale to global proportions. Autonomic systems contain feedback loops that provide continuous monitoring and, as needed, automatic reconfiguration of the systems. Autonomic systems have to provide the following features:
For this purpose Jade is built using a reflexive component model. Our current target application domain is that of clustered J2EE applications and Message-Oriented middleware. These applications rely on complex software (e.g. Apache, Tomcat, Jonas, Mysql, JMS server). We have made some experiments related to self-optimization, self-healing and self-protection with these use cases showing the interest of the approach.Slides (1.60 MB)
Many data modelling problems are naturally posed as existence of relations among observed variables rather than existence of a causal relation from one set of variables (inputs) to another set of variables (outputs). A consequence of this observation is that the original modelling problem is equivalent to the computational problem of low-rank approximation of a matrix constructed from the data rather than solution of an overdetermined system of equations. In this talk, I will present generic data modelling problems from signal processing, system theory, and machine learning that reduce to low-rank approximation and outline a few main approaches for solving the resulting (structured) low-rank approximation problems. The talk is based on the survey paper:
I. Markovsky, Structured low-rank approximation and its applications , Automatica, 44 pp. 891-909, 2008."
( web e-mail ), Doctoral candidate at Institut d'Administration des Entreprises, Lyon, France will give a talk:
Système d'acquisition vocale des connaissances sur le patrimoine culturel
Nous présentons un travail de recherche sur l'utilisation de l'interface vocale dans le processus de recueil et de modélisation des connaissances dans le domaine de la gestion du patrimoine culturel. Ce système hybride s'appuie sur des techniques de traitement du signal, de traitement du langage naturel et de modélisation des connaissances. Il décrit les étapes d'acquisition, d'extraction, et de modélisation des connaissances. Nous abordons également la problématique de la documentation du patrimoine culturel.Slides (962.18 kB)
We report our work on Word Sense Disambiguation (WSD) in a multilingual setting, with large scale applications in MT and Cross Lingual IR in view. The languages involved are
which are languages with very large speaker base. We rely on (i) domain specific dominant senses of words and (ii) the empirical but frequently made observation that within a domain the set of dominant senses remains the same across languages. The second fact allows us to perform disambiguation even in the absence of sense tagged corpora by using sense projections from annotated corpora of other languages. The proposed iterative, all-words disambiguation algorithm is greedy in nature and uses only previously disambiguated words as clues. It works on a graph of sense nodes erected column wise on the words of the sentence, as is done in PageRank based WSD. The function for scoring the senses is inspired by the energy expression in Hopfield network. It combines the "self-energy" of a sense node with the "interaction energy" with other sense nodes. The accuracy values of approximately 75% (F1-score) for all the languages in two different domains compares well with other state of the art algorithms like PageRank.Slides (999.94 kB)