Agenda

In this thesis, we focused on the representation of graphs through the problem of universal graphs. A graph is an abstraction used to represent the interconnections between different objects; these objects are called vertices, and their connections are called edges. It is common to represent graphs as a matrix or as a list of vertex adjacencies. These traditional representations form the basis of graph algorithms, which are essential in many fields such as telecommunications, electronics, and computer science. However, in certain specific contexts—such as distributed computing—constraints on space and the number of communications necessitate the use of more compact, so-called implicit, representations. Labeling schemes address this challenge. An adjacency labeling scheme is, for a family of graphs, an assignment of labels to the vertices of the graphs in the family such that, given a pair of labels, and without any other information, it is possible to determine whether the corresponding vertices are adjacent or not. The goal, then, is to minimize the size of these labels. We have focused on this problem more specifically through a related problem: that of induced universal graphs and their minimum number of vertices. An induced universal graph for a family of graphs contains, as induced subgraphs—that is, by selecting only a subset of the graph’s vertices and all the edges connecting them—the set of all graphs in the family. The main results of this thesis concern the impossibility of constructing universal induced graphs for certain families of graphs using fewer than a certain number of vertices. These fairly general results allow us to provide lower bounds for families based on their characteristics, such as the number of graphs in the family or the presence of unions of complete graphs of a certain size within the family. We also studied the number of graphs needed to improve these lower bounds; this was achieved using a construction of universal induced graphs for small families of graphs, in particular subfamilies of minor-closed graphs such as planar graphs. We also present graph constructions induced universal graphs for various families, such as star forests and unions of complete graphs. For the latter, the construction we propose is optimal in terms of the number of vertices. Finally, we present complexity results for induced universal graphs of minimal size and generalize these results to other types of universal graphs.

Cécile Gombert is giving this seminar—which is open to everyone—as part of the Numerics program.

The rapid growth of micromobility is gradually transforming urban travel patterns. Electric bicycles and scooters are now effective alternatives to traditional modes of transportation, helping to reduce traffic congestion and pollutant emissions. However, integrating these vulnerable users into cooperative intelligent transportation systems (C-ITS) poses new challenges in terms of connectivity, service continuity, mobility management, and communications security. Vehicle-to-Everything (V2X) communications are a key component of C-ITS architectures. They rely primarily on ITS-GS technologies, based on the IEEE 802.11p standard, and on next-generation cellular networks such as SG NR-V2X. Although these technologies have complementary characteristics, their effective deployment in micromobility environments remains a major challenge due to the dynamic mobility of users, the power constraints of onboard equipment, and the variability of radio conditions. This thesis focuses on evaluating and improving the performance of ITS communication technologies designed for micromobility. Initially, several representative use cases were studied in an ITS-GS-based VANET environment, including adaptive speed control, dynamic geofencing, and energy optimization of communication infrastructure. Intelligent management mechanisms for Roadside Units (RSUs), including the use of mobile RSUs, have been proposed to improve network coverage while reducing energy consumption. The results show a reduction in RSU energy consumption of up to 44.5%, while ensuring the continuity of V2X communications. Subsequently, a hybrid ITS-GS/SG architecture was developed to take advantage of the complementarity between low-latency local communications and wide-area cellular networks. An intelligent selection strategy a radio interface selection method, based on deep reinforcement learning and the Double Deep Q-Network (DDQN) algorithm, was proposed to dynamically optimize the choice of communication interface. This approach improves communication reliability, packet delivery rate, throughput, and energy efficiency in dynamic urban environments. Evaluations showed a packet delivery rate (PDR) of 99% and a 15.6% reduction in energy consumption.
Finally, this thesis addresses the issue of vertical handover in hybrid ITS-GS/SG architectures. An innovative approach to handover management, based on offline reinforcement learning and the Conservative Q-Learning (CQL) algorithm, is introduced. The proposed solution, called RIMA-HO, leverages historical data to reduce unnecessary transitions between radio technologies while ensuring service continuity and the quality of V2X communications. The RIMA-HO solution reduces unnecessary handovers and improves service continuity, with an average energy reduction of 16.3%, reaching 41.8% in the most favorable scenarios. The proposed contributions are validated through simulations conducted in a co-simulation environment integrating SUMO and OMNeT++, as well as using real-world mobility and communication traces. The results demonstrate significant improvements in connectivity, energy efficiency, communication stability, and mobility management. This thesis thus contributes to the development of intelligent solutions that promote the secure and efficient integration of micromobility into future cooperative transportation systems.

Clémence Frioux is giving this seminar—which is open to everyone—as part of the Numerics series.

Yassine Hamoudi is giving this seminar—which is open to everyone—as part of the Numerics series.

Camille Jeunet Kelway is giving this seminar—which is open to everyone—as part of the Numerics series.

Digital technology now plays a central role in our societies, but women remain significantly underrepresented in computer science education and careers. In most European countries, they account for less than 15% of students in computer science graduate programs. This situation raises not only a question of equality but also a democratic issue: it is difficult to imagine a digital world that truly serves everyone when its design, development, and maintenance rely on a predominantly male workforce.
This lecture aims to shed light on the mechanisms that have led to the gradual masculinization of computer science as the field has gained social and economic significance. It will then present the main challenges related to gender diversity in computer science programs and, more broadly, in mathematics. Finally, drawing on research conducted over the past twenty years on gender and technology, it will discuss various practices aimed at promoting the inclusion of women: some that are widespread but ineffective, and others that are more likely to produce lasting effects when they are part of a comprehensive institutional strategy.

 

Irénée Regnauld is giving this seminar—which is open to everyone—as part of the Numerics series.

Camille Castera is giving this seminar—which is open to everyone—as part of the Numerics series.

Jérôme Cieslak is giving this seminar—which is open to everyone—as part of the Numerics series.