In this talk, I’ll give a global overview of challenges related to the learning of dynamical models of cellular differentiation processes. I’ll present on-going work mixing formal methods for modelling dynamical systems and model synthesis from constraints on expected emerging behaviors. This will be illustrated in the scope of the synthesis and analysis of ensembles […]

The forecasted applications of machine learning (ML) in safety critical applications hinge on systems that are robust in their operation and that can be trusted. This talk overviews recent efforts on applying automated reasoning tools in explaining non-interpretable (black-box) ML models. Moreover, the talk details the computation of rigorous explanations of black-box models, and how […]

In the last years, data sciences have shown their interest (and somehow their limitation) in life sciences to extract information from multi-scale, incomplete, heterogeneous but somehow interdependent data. As an important bottleneck, figuring out how data sciences techniques can assist life sciences to provide explainable evidences of the limitations of current knowledge in biology and […]

Durant ce séminaire, nous nous intéressons à la fusion de deux domaines de recherche: la programmation par contraintes (PPC) et l’interprétation abstraite. La PPC est un paradigme déclaratif permettant de trouver les solutions d’un système de contraintes. La PPC s’applique à de nombreux domaines tels que l’ordonnancement, les tournées de véhicules ou encore en bioinformatique […]

We study voting rules for participatory budgeting, where a group of voters collectively decides which projects should be funded using a common budget. Many city governments around the world now use participatory budgeting to allow their residents to influence how the city spends its budget. Most cities handle such elections using the most intuitive voting […]

The variability of physical phenomena and partial ignorance about them motivated the development of probability theory in the two last centuries. However, the mathematical framework of probability theory, together with the Bayesian credo claiming the inevitability of unique probability measures for representing agents’ beliefs, have blurred the distinction between variability and ignorance. Modern theories of […]

Non-cooperative game theory is particularly useful for modeling interactions between competitive agents. In AI, it is the grounding framework of (non-cooperative) multi-agent planning, for example. AI has also contributed new representation frameworks (graphical games, hypergraphical games, Bayesian Action Graph Games…) and new solution algorithms (Multi-agent RL,…) to game theory. In this introductory talk, I will […]

In this talk, I will present some recent studies on social ranking from axiomatic and computational point of views and show it’s relation with power indices. In many domains, a number of works have been devoted to ranking individuals/objects based on the performance of the groups formed by them. Besides game theory, arguably the field […]

Ontology-mediated query answering (OMQA) is a promising approach to data access and integration that has been extensively studied by the knowledge representation & reasoning and database communities. Ontologies are used to enrich and unify the vocabulary of data sources, allowing users to formulate their queries in a more familiar vocabulary which abstracts from the specific […]

De Tesla à Waymo en passant par Stellantis, Renault, Mobileye et Valéo, la plupart des acteurs du monde de l’automobile revendiquent l’intégration de l’intelligence artificielle dans leurs produits. Mais de quelle IA parle-t-on ? A partir d’une vision historique, nous retracerons les principales étapes de l’introduction de l’IA dans les véhicules. Nous ferons ensuite un […]

State-of-the-art high-throughput biology techniques (such as single cell transcriptomics and live cell imaging microscopy) produce massive amounts of biological and clinical data, which consist of either independent snapshots or time series of observational data with just a few control parameters and experimental conditions. Yet, this wealth of heterogeneous data remain largely under-explored due to the […]

Writing software is tedious, error-prone, and accessible only to a small share of the population – yet coding grows increasingly important as the digital world plays larger and larger roles in peoples’ lives. Program synthesis seeks to make programming more reliable and accessible by developing methods for automatically constructing programs from specifications. In this talk, […]

Political debates offer a rare opportunity for citizens to compare the candidates’ positions on the most controversial topics of a political campaign. Thus they represent a natural application scenario for argumentation. On the one hand, the ability of identifying argumentative components and predicting their relations in such a kind of texts opens the door to […]

Event-condition-action (ECA) rules equip a database with information about preferred ways to repair it when the integrity constraints are violated. They face problems of non-terminating executions and only procedural semantics had been given to them up to now. Declarative semantics however exist for a particular class of ECA rules lacking the event argument, called active […]

Federated learning (FL) is a machine learning setting where many clients (e.g. mobile devices or whole organizations) collaboratively train a model while keeping their data decentralized. FL embodies the principles of focused data collection and minimization, and can mitigate many of the systemic privacy risks and costs resulting from traditional, centralized machine learning and data […]

Personal data are increasingly disseminated over the Web through mobile devices and smart environments, and are exploited for developing more and more sophisticated services and applications. All these advances come with serious risks for privacy breaches that may reveal private information wanted to remain undisclosed by data producers. It is therefore of utmost importance to […]

L’analyse de l’erreur humaine pour anticiper et prévenir les accidents reste un territoire peu exploré par l’IA. Pour cause : si le diagnostic de systèmes a fait l’objet de nombreux travaux, l’utilisation de modèles logiques pour diagnostiquer des erreurs humaines se heurtent à une difficulté majeure : l’erreur humaine n’est pas logique. Dans cet exposé, nous montrerons comment il est possible de modéliser, à l’aide de la logique, des phénomènes non-rationnels pour raisonner dessus à l’aide d’outils issus des méthodes formelles. Nous montrerons que cela nous a conduit à reconsidérer les problèmes classiques du décors, du diagnostic fondé sur la cohérence et de la révision de croyance.

Le choix social computationnel est un domaine interdisciplinaire à la croisée de l’économie, de l’IA et plus généralement de l’informatique. Il consiste à concevoir, analyser et calculer des mécanismes de prise de décisions collectives, avec divers sous-domaines d’application tels que le vote, l’allocation équitable des ressources, le budget participatif, la sélection de groupes d’individus représentatifs, ou l’appariement avec préférences. La rencontre de l’IA et du choix social n’a pas seulement permis de développer des algorithmes pour la prise de décision collective : elle a contribué à remodeler et à revitaliser le domaine, en identifiant de nouveaux paradigmes, de nouveaux problèmes et de nouveaux objets d’étude. Au cours de l’exposé, je donnerai quelques exemples montrant le rôle de diverses sous-communautés de l’IA dans l’étude de la prise de décision collective.

Si les travaux en « explicabilité » abondent, ils font rarement le lien avec ceux menés en théorie de la décision, qui fondent les diverses procédures appliquées sur un certain nombre de modèles de rationalité.
Dans cet exposé, nous montrerons qu’il est possible de proposer des moteurs d’explication des préférences fondés sur l’articulation de principes décisionnels. En enrichissant la palette d’approches disponibles, ces moteurs nous permettront en retour de mieux cerner les problématiques associées à la production d’explications.

A few years ago, François Chollet introduced a measure of intelligence, and a form of IQ test for machines, called the Abstraction and Reasoning Corpus (ARC). His motivation was to foster AI research towards more general and less task-specific artificial intelligence. He defines intelligence as skill acquisition efficiency rather than skill performance, where acquisition efficiency means learning with limited prior knowledge and few training examples. ARC is a set of tasks that consist in learning a transformation of small colored grids, from a few examples. Public challenges have shown the difficulty of ARC for machines compared to humans, despite its apparent simplicity. In this talk, after presenting Chollet’s measure of intelligence and ARC, I will present an approach that gives encouraging results on ARC, and that can be applied to other kinds of tasks. Existing approaches use program synthesis where programs are essentially function compositions, i.e. expressions. In contrast, we use object-centric models that combine patterns and expressions, and have a non-deterministic semantics. We also use the Minimum Description Length (MDL) principle in order to guide the search for models. This enables us to learn more complex models, in less time, and with less prior knowledge.

The Covid-19 health crisis has seen an increase in the use of digital work platforms from videoconferencing systems to MOOC-type educational platforms and crowdsourcing and freelancing marketplaces. These levers for sharing knowledge and learning constitute the premises of the future of work. Educational technologies coupled with AI hold the promise of helping learners and teachers. However, they are still limited in terms of social interactions, user experience and learning opportunities as they must address a tension between learner-centered and platform-centered goals. I will describe research at the intersection of data-informed recommendations and education theory and conclude with ethical considerations in building educational platforms.

Allocating multiple resources is an ubiquitous problem in multi-agent systems or economics with many applications such as allocating schools or universities to students, medical resources to hospitals, public housing to families… To promote acceptance of resource allocation systems, it is necessary to improve user confidence by developing reliable, fair and interpretable allocation mechanisms. Different fairness notions have been described in economics and computational social choice. In this talk, we will mainly focus on the notion of envy and argue that minimizing the envy is a natural criterion to promote acceptance of an allocation. We will review various allocation contexts (among social networks, under partial observability, with diversity constraints) and propose different relaxations of the envy-freeness notion. We will then investigate interpretable multi-agent protocols for fairly allocating resources and promoting acceptance of the outcome.

The theory of belief functions is a powerful formalism for uncertain reasoning, with many successful applications to knowledge representation, information fusion, and machine learning. Until now, however, most applications have been limited to problems (such as classification) in which the variables of interest take values in finite domains. Although belief functions can, in theory, be defined in infinite spaces, we lacked practical representations allowing us to manipulate and combine such belief functions. In this talk, I show that the theory of epistemic random fuzzy sets, an extension of Possibility and Dempster-Shafer theories, provides an appropriate framework for evidential reasoning in general spaces. In particular, I introduce Gaussian random fuzzy numbers and vectors, which generalize both Gaussian random variables and Gaussian possibility distributions. I then describe an application of this new formalism to nonlinear regression with uncertainty quantification.

https://www.youtube.com/watch?v=KMXs-viYVvI La morphologie mathématique est une théorie non linéaire d’analyse de structures qui a été largement appliquée à l’analyse d’images. Ses fondements mathématiques proviennent de l’algèbre, de la théorie des treillis complets ou encore de la topologie. Nous présentons dans cet exposé des liens entre cette théorie et la logique. Dans le cadre de la […]

Les fonctions de croyance constituent une classe de mesures qui sont suffisamment expressives pour capturer à la fois des connaissances probabilistes et des formes d’ignorance. Dans cet exposé, on parlera de résultat récents obtenus au sein du 3IA ANITI autour de problèmes de décision sous incertitude, et en particulier de problèmes de recherche d’équilibres de […]

L’exposé propose une introduction aux proportions analogiques et à leurs applications. Les proportions analogiques sont des énoncés de la forme "a est à b ce que c est à d”, où a, b, c, d peuvent référer à des mots, des phrases, des images, des structures, etc. On considère ici que a, b, c, d […]

Dans cet exposé, je présenterai nos efforts pour construire un agent socialement interactif doté de la capacité de montrer une large gamme de comportements communicatifs et émotionnels. Au cours d'une interaction, les humains adaptent continuellement leurs comportements à différents niveaux : formalité du langage, imitation, synchronisation, etc. En tant que partenaire d'une interaction, l'ISA doit […]

Dans cette présentation, nous abordons des modèles et techniques pour l’allocation de tâches dans un cadre multi-agents, comme dans des collectifs de robots autonomes. Dans ce contexte, un ensemble d’agents doit se coordonner pour décider de l’affectation d’une ou plusieurs tâches à chaque agent, en fonction du contexte opérationnel. Ce contexte peut inclure la position […]

In a decision-aiding process, to provide convincing recommendations that a decision-maker can fully understand and accept, a decision-aider must often engage in interaction and consider the decision-maker’s responses. This feedback can lead to revising the model used to represent the decision-maker’s preferences. Our objective in this work is to equip an artificial decision-aider with this […]

Tirer pleinement parti de la multiplicité des données disponibles aujourd’hui est un défi scientifique et technologique majeur, avec des impacts significatifs sur de nombreuses applications. Cet exposé s’inscrit dans ce contexte et aborde la question de savoir comment les méthodologies d’IA peuvent gérer des données hétérogènes de nature et de qualité différentes. Nous utiliserons des […]