2021


Retour à la vue des calendrier
Vendredi 12 Novembre
Heure: 12:30 - 13:30
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Relation Extraction with Distant Supervision: noise Reductio
Description: Juan Luis Garcia-Mendoza Distant Supervision is an approach that allows automatic labeling of instances. This approach has been used in Relation Extraction. Still, the main challenge of this task is handling instances with noisy labels (e.g., when two entities in a sentence are automatically labeled with an invalid relation). The approaches reported in the literature addressed this problem by employing noise-tolerant classifiers. However, if a noise reduction stage is introduced before the classification step, this increases the macro precision values or keep the same values with fewer instances. An approach based on Adversarial Autoencoders is proposed to obtain a new representation that allows noise reduction in Distant Supervision.
Jeudi 18 Novembre
Heure: 10:30 - 11:30
Lieu: ATTENTION Salle D215, bâtiment D, Université de Villetaneuse
Résumé: Optimization + Simulation: how to reduce bus bunching
Description: Yasmin A Rios Solis Real-time control strategies palliate with the day's dynamics in bus rapid transit systems. In this talk, we focus on a bus bunching problem that minimizes the number of buses of the same line cruising head-to-tail or arriving at a stop simultaneously by using bus holding times at the stops. For this, we propose a new mathematical model with quadratic constraints, whose objective function minimizes the penalties caused by buses that are bunching. Experimental results on a simulation of a bus rapid transit system in Monterrey, Mexico, show the efficiency of our approach. The results show a bus bunching reduction of 45% compared to the case without optimization. Moreover, in some scenarios the passenger waiting times are reduced by 30%.
Jeudi 25 Novembre
Heure: 10:30 - 11:30
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: SMS++: a Structured Modelling System with ... hopefully, one day ... some useful application?
Description: Antonio Frangioni On February this year, after about 8 years of gestation, an early beta release of the Structured Modelling System++ has been released to general public availability. SMS++ is a C++ library intended to facilitate the development of very large optimization problems with multiple nested heterogeneous structure, and especially of the corresponding solution methods, chiefly (but not exclusively) ones based on (parallel) decomposition. In the attempt of achieving this goal SMS++ has accrued a number of features that look quite unique in the landscape of modelling systems, so much so as to raise the legitimate suspicion that the reason why these features have never been developed before is because no sane person would have ever thought them a good idea. Yet the system is there and it does seem to offer some new viewpoints on mathematical modelling systems that may at least be worth a look. SMS++ is itself developed in an highly modular fashion and already counts a(n hopefully growing) number of separate sub-projects besides the "core" library and the support tools. One of these allows to solve Lagrangian Duals of complex integer programs with remarkable ease, and it will hopefully soon be joined by a similar component doing Benders' decomposition. Hence, there may actually be a few use cases in which SMS++ could be worth considering already, despite the very many missing components that would be needed to make it a really compelling prposition. In fact, perhaps the most interesting feature of SMS++ is it being community-oriented and (at least in principle) almost infinitely extendable to try to cater for the very diverse needs of the disparate clades of the optimization world. This alone may make it worth a second look, notwithstanding the arguably insane delusions of an all-conquering modelling system that some of the developers harbour and that would require capturing an unfeasibly large amount of mindshare to achieve.