2024


Retour à la vue des calendrier
Jeudi 18 Janvier
Heure: 10:30 - 11:30
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Rule-based machine learning via mathematical optimization
Description: Cristina Molero del Río Rule-based machine learning models are appealing because of their simple
decision structure. In this talk, we will present two examples, decision
trees and rule sets, with special focus on the former.

Contrary to classic classification and regression trees, built in a greedy
heuristic manner, designing the tree model through an optimization problem
allows us to easily include desirable properties in Machine Learning in
addition to prediction accuracy. We present a Non-Linear Optimization
approach that is scalable with respect to the size of the training sample,
and illustrate this flexibility to model several important issues in
Explainable and Fair Machine Learning. These include sparsity, as a proxy
for interpretability, by reducing the amount of information necessary to
predict well; fairness, by aiming to avoid predictions that discriminate
against sensitive features such as gender or race; the cost-sensitivity
for groups of individuals in which prediction errors are more critical,
such as patients of a disease, by ensuring an acceptable accuracy
performance for them; local explainability, where the goal is to identify
the predictor variables that have the largest impact on the individual
predictions; as well as data complexity in the form of observations of
functional nature. The performance of our approach is illustrated on real
and synthetic data sets
Jeudi 25 Janvier
Heure: 10:30 - 11:30
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: A branch-and-bound method for multiobjective mixed integer quadratic programs based on dual relaxations
Description: Marianna De Santis Most real-world optimization problems in the areas of applied sciences, engineering and economics involve multiple, often conflicting and nonlinear, goals. In the mathematical model of these problems, under the necessity of reflecting discrete quantities, logical relationships or decisions, integer and 0-1-variables need to be introduced, leading to MultiObjective Mixed Integer Nonlinear Programming problems (MO-MINLPs). The practical relevance of MO-MINLPs is pointed out in many publications, where tailored approaches for specific applications have been proposed. MO-MINLPs are intrinsically nonconvex, implying that the design of exact and efficient solution methods is particularly challenging and requires global optimization techniques. In this talk, we present a branch-and-bound method for multiobjective mixed-integer convex quadratic programs that computes a superset of efficient integer assignments and a coverage of the nondominated set. The method relies on outer approximations of the upper image set of continuous relaxations. These outer approximations are obtained addressing the dual formulations of specific subproblems where the values of certain integer variables are fixed. The devised pruning conditions and a tailored preprocessing phase allow a fast enumeration of the nodes. Despite the fact that we do not require any boundedness of the feasible set, we are able to prove that the method stops after having explored a finite number of nodes. Numerical experiments on instances with two, three, and four objectives are presente
Jeudi 8 Février
Heure: 10:30 - 11:30
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Exponentially large arc-flow models
Description: François Clautiaux Network flow formulations are among the most successful tools to solve optimization problems. Such formulations correspond to determining an optimal flow in a network. One particular class of network flow formulations is the arc flow, where variables represent flows on individual arcs of the network. In this talk, we will review classical and recent results on integer linear programming models based on arc-flow formulations in exponentially or pseudo-polynomial size networks. We will study the limitations of these approaches, and how various almost disconnected groups have addressed these limitations. We will describe a recent approach based on the generalization of these models to flow in hypergraphs, and propose some research directions.
Heure: 12:15 - 13:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: The role of Knowledge Graphs in externalizing information from conceptual models
Description: Ana-Maria Ghiran Due to the machine readable format used by Knowledge Graphs (KGs) in representing facts, and ontological models, they enabled AI systems to make decisions or to provide humans with insights by revealing hidden relationships between entities. Nevertheless, decision making in enterprises is far from being assigned to AI. Describing and evaluating business processes take the form of visual models that gained increased popularity among managers. But a business process diagram, usually described in the standardized notation BPMN (Business Process Model and Notation), enables more than just a visual representation of the knowledge – it creates a structured encoding of knowledge, which can be captured in a graph-based format. In this way, information that captures diverse facets of an enterprise (e.g. about business processes, resources, strategies, goals etc.) and that was mainly used by business executives and restricted to human interpretation, is externalized as KGs and provided for machine interpretation, thus enabling reasoning and semantic linking with external knowledge. In this presentation I will highlight that conceptual models should be considered as knowledge acquisition structures for any domain and that they can be processed as KGs with the help of Semantic Technology.
Lundi 12 Février
Heure: 12:15 - 13:00
Lieu: Salle B107, bâtiment B, Université de Villetaneuse
Résumé: Ethically-driven Multimodal Emotion Detection for Children with Autism
Description: Annanda Sousa Emotion detection (ED) aims to identify people’s emotions automatically. However, most ED
applications do not consider individuals who express emotions differently, such as people with
autism. Although studies have already focused on creating ED models tailored for children with
ASD, this application of ED suffers from a scarcity of resources and remains underperforming
compared to the state-of-the-art ED models for the general population.
This thesis addresses the gap in automatic ED between the general population and autistic
children while ensuring an ethically driven approach, i.e., having the well-being of participants
as the main priority during the whole research process.
To meet our research objectives, we created a data collection framework that minimises emo-
tional disruption to the participants, respects their privacy and rights according to GDPR, and
provides a dataset that can be shared with the research community. We created CALMED,
a multimodal annotated dataset for ED featuring children with autism that includes privacy-
preserving features, novel target emotion classes, annotations provided by the participants’ par-
ents and a researcher specialist who works with children with ASD.
Using the CALMED dataset, we created hundreds of models with unique configurations and
analysed them to explore the effectiveness of various methods for multimodal ED in autism.
Then, utilising the knowledge acquired in this analysis, we proposed a multimodal ED model
that outperformed the previous state-of-the-art, reaching 81.56% and 75.47% for accuracy and
balanced accuracy, respectively.
Finally, we created and shared many systems to support the data acquisition process and
data experiments creation and analysis. We placed great importance on ensuring reproducibility,
reusability, and ethical conduct.
This research has made significant contributions to the field of ED applied to ASD. It has
provided a valuable dataset, analytical insights, a state-of-the-art model, and many computer
systems that can serve as a groundwork for future work.
Jeudi 29 Février
Heure: 10:30 - 11:30
Lieu: https://bbb.lipn.univ-paris13.fr/b/wol-ma9-vjn
Résumé: Efficacité et équité dans le problème d'ordonnacement multi-organisation
Description: Martin Durand On considère le problème d'ordonnancement multi-organisation (POMO). Un ensemble de N organisations possèdent chacune un ensemble de machines et de tâches. Chacune de ses organisations dispose d'un ordonnancement, dit local, dans lequel elle ordonnance ses tâches sur ses machines. Notre but est de trouver un ordonnancement de toutes les tâches sur toutes les machines et tel que chaque organisation soit au moins aussi satisfaite dans cette solution globale qu'avec son ordonnancement local, cette contrainte est appelée contrainte de rationalité. On montre que la coopération peut permettre à toutes les organisations d'obtenir simultanément une meilleure solution. On étudie egalement à quel point la contrainte de rationalité impacte la qualité de la solution globale. Dans un second temps, on introduit un nouveau problème centré sur l'équité: on formule le bénéfice qu'une organisation obtient en coopérant et on étudie le problème de maximisation du plus petit bénéfice. On montre que ce problème est fortement NP-difficile et inapproximable dans le cas général et on propose une heuristique polynomiale qui retourne de bonnes solutions dans nos expérimentations.