ED Mathématiques et Informatique
Integration of preferences and domain knowledge in inconsistency-tolerant query answering
by Robin JEAN (LaBRI - Laboratoire Bordelais de Recherche en Informatique)
The defense will take place at 14h00 - 050-AMPHI 351, cours de la Libération F-33405 Talence Batiment A30
in front of the jury composed of
- Meghyn BIENVENU - Directrice de recherche - Université de Bordeaux - Directeur de these
- Vanina MARTINEZ - Chargée de recherche - Artificial Intelligence Research Institute (IIIA) - Examinateur
- Riccardo ROSATI - Professeur - Sapienza Università di Roma - Rapporteur
- Thomas EITER - Professeur - TU Wien - Rapporteur
- Salem BENFERHAT - Professeur des universités - Université d'Artois - Examinateur
In ontology-mediated query answering, an ontology (often specified using description logics, DLs) is used to enrich incomplete data with domain knowledge, which is taken into account when computing query answers. A central issue in this setting is how to proceed when the data is inconsistent with the ontology, since in classical logic an inconsistent theory entails every formula, so that inconsistency-tolerant semantics are needed to obtain meaningful answers. We consider three existing inconsistency-tolerant semantics based upon the notion of a repair, defined as an inclusion-maximal subset of the data consistent with the ontology. These three semantics can be used conjointly to identify answers with different levels of confidence: those that hold in every repair (AR semantics), those entailed by the facts common to all repairs (IAR semantics), and those holding in at least one repair (brave semantics). The thesis focuses on repair-based and cost-based semantics which take into account preference information reflecting the differing reliability of the facts and/or axioms, given either by a qualitative priority relation between conflicting facts or numerical weights. The thesis develops three complementary contributions aimed at bringing inconsistency-tolerant reasoning with preferences closer to practice. Our first contribution is a new cost-based semantics for DL knowledge bases, in which every ontology axiom and data assertion carries a weight reflecting its reliability, hard statements receiving infinite weight and soft ones finite weights. The cost of an interpretation aggregates the weight of the statements it violates, and certain and possible answers are then computed over the interpretations whose cost is either below a threshold or the minimum. For this setting, we analyze the complexity of the relevant reasoning problems, covering DLs from the lightweight EL family to the more expressive ALCO, instance and conjunctive queries, and both combined and data complexity. The second contribution concerns preference specification, where we introduce a general rule-based method for modelling preferences over conflicting facts and deriving the priority relation that prioritized repair-based semantics require but that earlier work simply assumed to be given. A novel aspect of our work is the treatment of cycles in the expressed preferences, handled both by statically deciding whether a set of rules is guaranteed to yield an acyclic relation and by offering pragmatic cycle-removal strategies, the whole approach being implemented in answer set programming. The third contribution is the implementation of query answering over prioritized data under the Pareto-, globally-, and completion-optimal repair-based semantics, using answer set programming and its quantified extension ASP(Q). To our knowledge, ours is the first implementation to support the globally-optimal repair-based semantics. The approach is based on shared building blocks, a localization technique that restricts reasoning to the relevant part of the data, and tractable approximations, in particular based on the grounded repair. An empirical study indicates that although reasoning over inconsistent data enriched with preferences is theoretically hard, it can often be handled in practice by exploiting these tractable approximations and the capabilities of modern ASP solvers. While formulated for DL knowledge bases, the second and third contributions also apply to other ontology and database constraint frameworks.
ED Sciences de la Vie et de la Santé
Beyond satiety: deciphering the role of hypothalamic POMC neurons in driving the consumption of Hypercaloric food
by Victor JOUQUE (Neurocentre Magendie)
The defense will take place at 13h00 - Amphi Centre Broca Nouvelle - Aquitaine 146 Rue Léo saignat 33000 Bordeaux
in front of the jury composed of
- Daniela COTA - Directrice de recherche - Neurocentre Magendie INSERM U1215 - Directeur de these
- Carole ROVERE-JOVENE - Chargée de recherche - Institut de pharmacologie moléculaire et cellulaire, Université Côte d'Azur - Rapporteur
- Amandine GAUTIER-STEIN - Directrice de recherche - laboratoire Nutrition, Diabète et cerveau, U1213, INRAE - Rapporteur
- Giuseppe GANGAROSSA - Professeur des universités - Unité de Biologie Fonctionnelle et Adaptative (CNRS UMR 8251) Université Paris-Cité - Examinateur
- Serge LUQUET - Directeur de recherche - Unité de Biologie Fonctionnelle et Adaptative (CNRS UMR 8251) Université Paris-Cité - Examinateur
Introduction: Eating is a process essential for life. Therefore, powerful brain mechanisms have evolved to allow not only the matching of the organism's energy needs with energy intake, but also the recognition of food rich in calories so as to guarantee survival under variable environmental food sources. In this context, neuronal circuits classically aiming at integrating information about the organism's energy status must interact with networks regulating the rewarding aspect of food intake. Although exposure to calorie-rich diets is well known to promote overeating and consequent obesity in today's modern societies, little is known about how such communication is set in place and the neuronal substrates underlying this phenomenon. Hypothalamic pro-opiomelanocortin (POMC) neurons are classically viewed as the mediators of satiety via the release of the neuropeptide α-MSH in response to metabolic and hormonal cues. However, recent evidence demonstrates that POMC neurons are highly heterogeneous, can become active before food consumption and stimulate feeding under specific conditions. In addition, POMC neurons release the µ-opioid receptor (MOR) agonist β-endorphin, known to stimulate the intake of energy-dense food, therefore challenging the traditional view of their satietogenic function. Objectives: The general aim of this thesis was to study the role of hypothalamic POMC neurons in the consumption of energy-dense palatable food. The specific objectives were the following: - (1) Define the mechanism engaged by POMC neurons in the hyperphagia toward hypercaloric food; - (2) Unravel the neuronal circuit implicated Results: By combining neuroanatomical, electrophysiological and in vivo calcium imaging approaches at single cell resolution, we demonstrate that POMC neurons are rapidly activated by consumption of high-fat diet (HFD) through orosensory processes. Activation of hypothalamic POMC neurons is specific to HFD exposure and it drives hyperphagia in both sexes, as chemogenetic inhibition of POMC neurons reduces HFD intake. These effects seem to require the synthesis and release of the POMC-derived peptide β-endorphin, as its levels increase in POMC terminals upon HFD exposure, while HFD-driven hyperphagia can be selectively inhibited by MOR antagonism. Single nucleus RNA sequencing (SnRNAseq) data further show that POMC neurons activated in response to HFD have increased expression of molecular mechanisms controlling the synthesis, addressing and release of β-endorphin. Using anterograde tracing of POMC neuronal projections, we then reveal that POMC neurons project to several structures implicated in food reward, especially the lateral septum. Using patch clamp recording coupled with optogenetics, we demonstrate that POMC neurons are electrically connected with LS neurons through an inhibitory neurocircuit. Pharmacological inhibition of MOR activity specifically in the LS, but not in other brain structures receiving POMC neuron projections, recapitulates the hypophagic effects induced by general MOR blockade. Finally, in vivo optogenetic inhibition of the POMC-LS neurocircuit reduces HFD intake, demonstrating that the lateral septum acts as a downstream brain region mediating HFD-driven hyperphagia by integrating POMC-derived β-endorphin. Conclusion: Our work suggests that, contrary to established dogma, certain POMC neurons subpopulations drive the intake of hypercaloric and palatable food through a previously uncharacterized neurocircuit involving the release of β-endorphin in the lateral septum. These set of findings, altogether provide novel information on the complex role played by hypothalamic POMC neurons in the regulation of feeding beyond satiety.