2016


Retour à la vue des calendrier
Lundi 2 Mai
Heure: 14:00 - 15:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Neoveille, état d'avancement du projet
Description: Emmanuel Cartier Neoveille est un projet financé par SPC qui vise à construire une plateforme de repérage, d'analyse et de suivi des néologismes sur gros corpus, en sept langues (français, portugais du Brésil, tchèque, grec, polonais, russe, chinois).
Nous rappellerons les objectifs globaux du projet avant d'évoquer les différents composants de la plateforme dans le détail : récupération automatique de corpus en sept langues; repérage automatique des néologismes de forme; indexation dans le moteur de recherche et d'analyse.
Finalement, nous évoquerons les pistes de travail à venir.
Lundi 9 Mai
Heure: 14:00 - 15:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Motifs pour la caractérisation des genres textuels et des styles
Description: Dominique Legallois
Mardi 10 Mai
Heure: 14:00 - 17:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Partially ordered sets
Description: Henri Mühle
Jeudi 12 Mai
Heure: 12:15 - 13:30
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Agrégation souple et adaptative des graphes hétérogènes avec des attributs hétérogènes
Description: Amine Louati In the enterprise context, people need to exploit, interpret and mainly visualize dierent types of interactions between heterogeneous objects. Graph model is an appropriate way to represent those interactions. Nodes represent the individuals or objects and edges represent the relationships between them. However, extracted graphs are in general heterogeneous (i.e., composed of different node attributes and different relationship types) and large sized which makes it diffcult to visualize and to analyze easily. An adaptive aggregation operation is needed to have more understandable graphs in order to allow users discovering underlyin g information and hidden relationships between objects. Existing graph summarization approaches such as k-SNAP are carried out in homogeneous graphs where nodes are described by the same list of attributes that represent only one community. The aim o f this work is to propose a general tool for graph aggregation which addresses both homogeneous and heterogeneous graphs. To do that, we develop a new soft and adaptive approach to aggregate heterogeneous graphs using the definition of Rough Set Theory (RST) combined with Formal Concept Analysis (FCA), the well known K-Medoids and the hierarchical clustering methods. Aggregated graphs are produced according to user-selected node attributes and relationships. To evaluate the quality of the obtained summaries, we propose two quality measures that evaluate respectively the similarity and the separability of groups based on the notion of common neighbor nodes. Experimental results demonstrate that our approach is effective for its ability to produce a high quality solution with relevant interpretations.
Mardi 17 Mai
Heure: 14:00 - 17:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Analytic combinatorics
Description: Bernhard Gittenberger
Mercredi 18 Mai
Heure: 14:00 - 15:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Vers la conception formelle de systèmes d’édition collaborative consistants
Description: Hanifa Boucheneb Les systèmes d’édition collaborative permettent à un groupe d’utilisateurs de partager et modifier des objets (textes, images, d ocuments XML, etc.) via le Web. Pour une meilleure réactivité aux opérations d’édition, ces systèmes sont en général basés sur la réplication des données. Chaque utilisateur a donc sa propre copie (locale) de l’objet qu’il peut modifier. Les modifications (opérations) locales sont ensuite propagées et intégrées aux autres copies. Un des défis majeurs de ces systèmes est d’assurer la consistance des données répliquées. Ce séminaire présentera et discutera les principales approches d’intégration des modifications non locales, proposées dans littérature. Il considérera ensuite les approches basées sur la transformation des opérations (OT) et montrera comment utiliser les méthodes formelles (model-checking symbolique et synthèse de contrôleur) pour vérifier si une approche OT assure la consistance des données répliquées et synthétiser une approche OT qui assure la consistance des données répliquées.
Lundi 23 Mai
Heure: 14:00 - 15:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Modèles neuronaux pour la traduction automatique
Description: Alexandre Allauzen Les modèles neuronaux occupent aujourd'hui dans le traitement
automatique des langues (TAL) une place importante car ils permettent
grâce à leur caractère continu des avancées significatives dans de
nombreux domaines applicatifs. Historiquement, les modèles de langue
neuronaux ont été une des premières réalisations marquantes, avec des
applications en reconnaissance automatique de la parole (RAP), puis à
d'autres tâches complexes de modélisation linguistique, comme par
exemple l'analyse syntaxique, l'estimation de similarité sémantique, les
modèles d'alignement de mots et en traduction automatique statistique.
L'exposé décrira les travaux menés au LIMSI-CNRS sur les réseaux
neuronaux appliqués principalement à la traduction automatique: les
modèles de langues n-grammes à grand vocabulaire, puis leur extension
aux modèles de traduction et leur apprentissage discriminant.
Mardi 24 Mai
Heure: 14:00 - 17:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Physical models and Tracy-Widom
Description: Peter Nejjar
Mercredi 25 Mai
Heure: 14:00 - 15:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: On the reconstruction of trees from their U-polynomial.
Description: José Aliste-Prieto The U-polynomial of a graph was introduced by Noble and Welsh as a generalization of some invariants coming from Knot theory. It also generalizes the chromatic symmetric function of Stanley. In this talk, we will consider the problem of whether there exist non-isomorphic trees
with the same U-polynomial (or,equivalently, with the same chromatic symmetric function).

We will survey what is know about the U-polynomial and this problem. In particular, we will show how to recover some classic invariants from the U-polynomial and we exhibit several subclasses of trees for which a solution of this problem is known. FInally, we construct some non-isomorphic trees with "almost" the same U-polynomial, based on solutions of an old problem in Number theory due to Prouhet-Tarry-Escott.
Mardi 31 Mai
Heure: 14:00 - 17:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Counting connected graphs with large excess
Description: Élie de Panafieu We enumerate the connected graphs that contain a linear number of edges with respect to the number of vertices. So far, only first term of the asymptotics and a bound on the error were known (Bender Canfield McKay 1995, Pittel Wormald 2005, van der Hofstad Spencer2006). We present a proof based on analytic combinatorics, i.e. generating function manipulations, and derive the complete asymptotic expansion.
Lundi 20 Juin
Heure: 14:00 - 15:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Towards narrative generation of spatial experiences in service robots
Description: Jorge Garcia Flores We propose a first approximation to the generation of narrative
experiences by a service robot. The goal of such narratives is to
communicate in a brief, structured and natural way what the robot
recently experienced while solving a task. We structure the narrative's
elements into blocks guided by the spatial areas the robot visited. We
also include a mechanism to mention failures, and template-based natural
language generation. At this state of our research we include movement,
manipulation and visual events. We evaluate our approach comparing the
generated narratives with descriptions produced by humans after having
followed the same robot's trajectory and performed the same robot's
activities. We use the ROUGUE-L automatic evaluation metric and a survey
in order to measure the narrative quality of the robot's description.
Results show that while the robot generated narratives have close to
human completeness scores, they still lack of readability and fluency.
Lundi 11 Juillet
Heure: 14:00 - 15:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Abstraction, Entropy and Computing Formats
Description: Luis A. Pineda In this talk a theory for the diagrammatic representation and computation of
finite discrete functions and abstractions is presented. The theory is defined in
terms of two basic operations that are computed directly on tables: the func-
tional abstrac tion and the functional application or reduction. However, unlike
the analogous operations of the lamba-calculus, these operations are not fully re-
versible and the system has an inherent information loss. For this, abstractions
have an associated entropy value that measures their degree of indeterminacy
or information content. The theory is applied to the definition and construction
of an associative memory, where the information is accessed by content, with
its associated memory register, recognition and retrieval operations. A case
study in visual memory with very promising preliminary results is presented.
The overall theory suggests a comprehensive view or space of possible compu-
tations that is defined in relation to (1) the trade-off between extensional and
intensional forms of expressing information and (2) the formats employed in
computations. This trade-off underlies the knowledge representation trade-off
of articial intelligence and cognitive science.The computing formats, in
turn, range from the linguistic format, whose paradigmatic case is the Turing
Machine, to fully distributed formats including neural networks and the
diagrammatic format.The view suggests that the trade-off between extensions
and intensions is the manner in which the entropy of abstractions surface in the
linguistic format. It also supports the case of direct representation in AI and
the case of images in the imagery debate, and helps to clarify the opposition between
symbolic and sub-symbolic computations. Finally, the implications of the view for learning,
creativity, embodied and situated cognition, and for the distinction between "artificial" and
"natural” computations are briefly discussed.