- 14h : Guillaume Moinard (ComplexNetworks): « Flocking and Be Water: Robust tactics for Protests on Street Networks »

Consider the following scenario. Protesters are scattered throughout a city and want to gather into groups large enough to perform significant actions. They face forces that may break up groups, block some places or streets and seize any communication devices protesters may be carrying. As a consequence, protesters only have access to local information on people and streets around them. Furthermore, formed protester groups must keep moving to avoid containment by adversary forces.

In this scenario, protesters need a distributed and as simple as possible protocol, that utilises local information exclusively and ensures the rapid formation of significantly large, mobile, and robust groups. We present simple models that relies on random walkers on a street network that follow tactics built from sets of basic rules. In this talk, we will present the most efficient rules for fast and robust flocking of walkers.

- 14h35 : Abdelrahman Abdelazim (CIAN): « Design of an AI hardware accelerator for RF signal classification »

This research focuses on designing a lightweight, low-energy AI hardware accelerator for edge devices, enabling the execution of AI algorithms in the physical layer of wireless communication systems to enhance privacy, energy efficiency, and speed. Monitoring I/Q signals in physical layer allows for real time and early detection of anomalies and malicious attacks, specifically targeting applications such as signal identification and the detection of various malicious activities, including jamming attacks and covert channels.

A key motivation for this work is that most existing AI accelerators are primarily designed for image classification tasks, which are not optimized for over-the-air signals in wireless communication. Consequently, developing a dedicated accelerator for RF signal classification would significantly improve efficiency in terms of power consumption, area requirements, and processing speed.

- 15h10 : Jordan Thieyre (SMA): « Reassessing the Impact of Reading Behaviour in Online Debates Under the Lens of Gradual Semantics » While it is unrealistic to assume users of online debate platforms to read and interpret all the arguments available, it is important to understand how positions will emerge on the basis of a fraction of those arguments. What arguments exactly will be accessed by users depend on assumptions on the platform design or on the readers' behaviours. Young et al. were the first to explore this question and report results in the context of an underlying extension-based semantics. We undertake a similar study in the context of gradual semantics, using a more comprehensive set of metrics, testing a larger number of behaviours, and come to different conclusions. We show in particular that a reading behaviour balancing supports and attacks provides interesting results.

- 15h45 : petite pause

- 15h55 : Léo Nebel (MOCAH) « Investigating how changes in students’ revising behavior relate to the evolution of essay content quality » A lot of tools made to automatically assess writings emerged in the past few years and they are getting more and more accurate. However, these tools often focus on the writing product and do not evaluate the writing process, even if it has been shown that the process followed influences the text quality in most educational contexts. We are currently investigating the different behaviors of students in their writing processes through an automatic writing evaluation tool to explore the links between the writing processes and the quality of the writing product.

- 16h30 : Margot Hérin (Décision): « Noise-tolerant Active Preference Learning for Multicriteria Choice Problems »

To make a choice in the presence of multiple criteria, we generally use an aggregation function which determines, for each alternative, the balance of its strengths and weaknesses and its overall evaluation. The aggregation function uses weights to adapt the model to the decision-maker's value system, by specifying the importance of the criteria and possibly their interactions. In this paper, we propose a noise-tolerant active learning method for these parameters, which not only effectively reduces the indeterminacy of the weights to identify an optimal or near-optimal decision among a given set of alternatives, but also simultaneously determines a predictive model of preferences capable of making relevant choices for the decision-maker on new instances. These outcomes are achieved by leveraging a general disagreement-based active learning approach that is theoretically guaranteed to be tolerant to noisy answers. The proposed method applies to various weighted aggregation functions, linear or not, classically used in decision theory.

- 17h05 : Guillaume Gervois (LFI/SMA): « Defining compatibility for moral preferences »

En éthique computationnelle, on propose une traduction des principes éthiques dans un langage formel afin d'automatiser le raisonnement moral. Ces principes nous fournissent alors des préférences morales sur un ensemble de décisions réalisables. Il existe de nombreux principes éthiques et l'utilisation conjointe de différents principes au sein d'une unique procédure de décision nécessite une compréhension fine de la compatibilité des principes dans toutes les situations possibles du contexte d'application de la procédure de décision. Nous proposons une définition de la compatibilité pour des ensembles de préférences dans une situation donnée, ainsi qu'une représentation graphique des préférences qui soit lisible pour un humain en cas de compatibilité. Ce travail est une première étape vers la conception de méthodes d'évaluation des procédures de décisions faisant intervenir différents principes éthiques.

- 17h40 : Goûter

Une seconde journée aura lieu au printemps pour permettre à d’autres doctorants de présenter leurs travaux.