2018


Retour à la vue des calendrier
Lundi 3 Septembre
Heure: 16:00 - 17:30
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Analysing large-scale Research Data with Semantic Technologies
Description: Francesco OSBORNE Semantic Technologies provide useful solutions for the analysis of big scholarly data since they facilitate the integration of large datasets and support tasks such as Natural Language Processing and Information Retrieval. In particular, ontologies that describe research topics and their relationships proved to be effective tools for making sense of research dynamics, classifying publications, detecting research communities, and forecasting research trends. However, these knowledge bases are very expensive to craft and tend to become obsolete fairly quickly.
In my talk, I will discuss the automatic approach that we designed to generate and update the Computer Science Ontology (CSO), a large-scale ontology of research topics including about 25K concepts. CSO has been used for supporting Springer Nature in classifying editorial products, informing marketing decisions, and evolving their internal taxonomy. I will present some systems adopting this knowledge base and describe their effect on the workflow of a major publishing company. I will also discuss the advantage of combining Machine Learning and Semantic Technologies for addressing complex tasks such as predicting research trends and forecasting technology migrations.
Jeudi 13 Septembre
Heure: 10:15 - 11:30
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: CoGITARe
Description: Flavien Breuvart Mon ARNJCJC CoGITARe a été accepté cet été. Ce séminaire a pour but de présenter le projet, ses objectifs et ses différentes facettes en essayant de rester haut niveau.
Jeudi 20 Septembre
Heure: 10:15 - 11:30
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: GdT : SL=L
Description: Paulin groupe de travail sur le résultat d'égalité de classes SL=L

(Annonce très tardive car oubliée, désolé...)
Lundi 24 Septembre
Heure: 15:00 - 16:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: A UML-based Proposal for IoT System Requirements Specification
Description: Gianna Reggio The talk presents a preliminary version of IotReq, a method for the elicitation and specification of the requirements for an IoT system.

The first task suggested by IotReq is the modelling of the domain, using the UML and following the service-oriente d paradigm, then the goals of the IoT system to build are elicited and specified, again using the UML and extending the domain model, producing a specification of the functional requirements.

IotReq also provides preliminary indications for specifying the technological nonfunctional requirements.

A case study, the specication of the requirements for a system to support the Genoa’s Science Festival is presented too.
Mardi 25 Septembre
Heure: 12:30 - 13:30
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Modèles d'optimisation pour le transport urbain "flexible"
Description: Yasmin A. Rios-Solis Je présenterai des problèmes de transport urbain (bus) qui se trouvent
dans un contexte “flexible” ou plutôt, “désordonné”:
chauffeurs absents, événements inattendus qui arrivent aux véhicules,
pas de voie spéciale pour les bus, presque pas d’arrêts et
compétition entre compagnies de transport. Pour ces problèmes, il faut
de nouveau modèles d’optimisation, des preuves de complexité et de
nouvelles méthodes de résolution. Au lieu de rentrer dans les détails
mathématiques, je présenterai plutôt les thèmes de recherche qui
émergent de ce contexte flexible.
Heure: 12:30 - 15:30
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: GADTs
Description: Flavien Il s'agit d'une présentation général des GADTs par Flavien, avec une explication de leurs intérêts pratique. Cela sera suivit d'un cas d'étude présenté par Jean-Vincent sur lequel on pourra réfléchir tous ensemble. Le créneau est long, car on ira manger entre les deux parties.
Jeudi 27 Septembre
Heure: 15:30 - 17:00
Lieu: Salle A303, bâtiment A, Université de Villetaneuse
Résumé: Arrows
Description: Exequiel Rivas Using monads for structuring computational effects had a big impact in the functional programming community. Arrows (Hughes 2000) are a generalisation of monads which provides support for computational effects that may be partially static. This talk will begin with an introduction on the subject, and then continue to discuss one the possible semantics of arrows from a categorical perspective.
Lundi 1 Octobre
Heure: 15:00 - 16:00
Lieu: Salle A303, bâtiment A, Université de Villetaneuse
Résumé: One-Sided Communications for more Efficient Parallel State Space Exploration over RDMA Clusters
Description: Sami Evangelista This talk presetns the use of one-sided communications in the context of state space exploration. This operation is often the core component of model checking tools that explores a system state
space to look for behaviours deviating from its specification. It basically consists in the exploration of a (usually huge) directed graph whose nodes and edges represent respectively system states and system changes. We revisit the state of the art distributed algorithm and adapt it to RDMA clusters with an implementation over the OpenSHMEM library and report on preliminary experiments conducted on the Grid'5000 cluster. This asynchronous approach thus reduces the significant communication costs induced by process synchronisation in two-sided communications.
Mercredi 3 Octobre
Heure: 14:00 - 15:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: First Order Algorithms for Constrained Optimization Problems in Machine Learning
Description: Francesco Rinaldi Thanks to the advent of the "Big Data era", simple iterative first-order optimization approaches for constrained convex optimization have re-gained popularity in the last few years. In the talk, we first review a few classic methods (i.e., conditional and projected gradient method) in the context of Big Data applications. Then, we discuss both theoretical and computational aspects of some new active-set variants for those classic methods. Finally, we examine current challenges and future research perspectives.

DISCLAIMER: This aimes to be a wide audience talk (for any LIPN member, Ph. D. students included) and you are not assume to know what is a "first-order optimization approach", a "conditional or projected gradient method" or an "active-set variant".
Jeudi 25 Octobre
Heure: 12:15 - 13:30
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Apprentissage automatique et adaptatif pour le clustering de flux de données relationnelles
Description: Parisa Rastin Les approches basées sur les prototypes sont très populaires en apprentissage non supervisé, en raison de la compacité du modèle résultant (les prototypes), de la puissance descriptive de ces prototypes et de la faible complexité de calcul du modèle (chaque objet est comparé à un petit nombre de prototypes). Nous proposons une approche de K-moyennes relationnelle utilisant un ensemble unique de points de support à travers le processus d'apprentissage, puis on introduit le formalisme des Coordonnées Barycentriques afin d'unifier la représentation des objets et des prototypes, ce qui permet un processus d'apprentissage incrémental simple pour le clustering relationnel. Notre motivation pratique est de réaliser un profilage en temps réel des utilisateurs connectés. Les tâches de profilage visent à reconnaître "l'état d'esprit" des utilisateurs à travers leur navigation sur différents sites en ligne.
Lundi 5 Novembre
Heure: 14:00 - 15:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Learning to Navigate and Extract Information from Web Results
Description: Ivan Vladimir MEZA In this talk I present our progress into the ECOS-NORD project between LIPN-Paris 13 and IIMAS-UNAM at Mexico. The talk focus on our experimental setup that allows to learn to navigate web results and to extract information from them. In particular, at this stage we are focus on extracting biográfical information of researchers, in order to quantify the Mexican returning diaspora. Our experimental setup is based on a reinforcement learning setup, we use a labelled data to learn the main actions on the results grid. We will show preliminary results, and newlines of experimentation.
Heure: 15:00 - 16:00
Lieu: Salle A303, bâtiment A, Université de Villetaneuse
Résumé: State Compression Based on One-Sided Communications for Distributed Model Checking
Description: Laure Petrucci We propose a distributed implementation of the collapse compression technique used by explicit state model checkers to reduce memory usage. This adapatation makes use of lock-free distributed hash tables based on one-sided communication primitives provided by libraries such as OpenSHMEM. We implemented this technique in the distributed version of the model checker Helena. We report on experiments performed on the Grid'5000 cluster with an implementation over OpenMPI. These reveal that, for some models, this distributed implementation can altogether preserve the memory reduction provided by collapse compression and reduce execution times by allowing the exchanges of compressed states between processes.
Mardi 13 Novembre
Heure: 12:30 - 13:30
Lieu: Salle A303, bâtiment A, Université de Villetaneuse
Résumé: On importance splitting and the automation of rare event simulation
Description: Carlos E. Budde In the analysis of formal models, simulation based approaches like statistical model checking offer a solution to the state space explosion that hinders verification, but may suffer from long execution times.
This is exacerbated when the property value to approximate depends on an event seldom observed; in those cases rare event simulation (RES) techniques can speed up convergence by reducing the variance of the statistical estimator.
However, RES techniques typically require non-trivial and domain-specific (or even model-specific) user input, which is a setback w.r.t. the push-button approach of standard model checking.
In this talk I will briefly discuss methods to automate the implementation of a specific approach to RES called "importance splitting."
I will overview some known implementations and discuss two algorithms recently developed to select "importance thresholds" and "splitting/effort factors," which are parameters with direct impact on the efficiency of importance splitting.
A good performance of the outcomes of the algorithms proposed has been empirically demonstrated on several case studies.
Jeudi 22 Novembre
Heure: 10:30 - 12:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Soft modalities as prices: a game model for intuitionistic linear logic with subexponentials
Description: Carlos Olarte We look at substructural calculi from a game semantic point of view,
guided by certain intuitions about resource conscious and, more
specifically, cost conscious reasoning. To this aim, we start with a
game for aILL (affine intuitionistic linear logic), where player I
defends a claim corresponding to a (single-conclusion) sequent, while
player II tries to refute that claim. Branching rules for additive
connectives are modeled by choices of II, while branching for
multiplicative connectives leads to splitting the game into parallel
subgames, all of which have to be won by player I to succeed. The game
comes into full swing by using subexponentials for representing two
types of options - volatile and permanent - for purchasing
resources. This leads to a new type of subexponetial calculus where
costs are attached to sequents. Different proofs are interpreted as
more or less expensive strategies to obtain a certain resource from a
bunch of resources (priced options). Finally, we generalize the
concept of costs and option's prices in proofs by using a semiring
structure. This general framework allows us to
interpret a wider range of subexponential systems and give meaning
to the use of resources in proofs in a more flexible way.
Jeudi 29 Novembre
Heure: 10:15 - 12:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Linear MSO and Church synthesis
Description: Pierre Pradic