2018

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.

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".

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.

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.

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.

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.

Heure:	10:15 - 12:00
Lieu:	Salle B107, bâtiment B, Université de Villetaneuse
Résumé:	Linear MSO and Church synthesis
Description:	Pierre Pradic

Heure:	12:15 - 13:30
Lieu:	Salle B107, bâtiment B, Université de Villetaneuse
Résumé:	Integer Programming approaches for Vertex Coloring Problems and for some applied projects
Description:	Diego Delle Donne This seminar presents several variants of the classical vertex coloring problem together with an overview on some theoretical and practical results related to Integer Programming approaches for these problems. The practical aspect involves implementations of competitive algorithms for these problems while the theoretical aspect relates to polyhedral studies of some formulations with the goal of finding "nice" characterizations for the associ ated polytopes.