ED Mathématiques et Informatique
Humanoid Robot Locomotion, Footstep Planning and Fall Resilience via Reinforcement LearningPolicies
by Clément GASPARD (LaBRI - Laboratoire Bordelais de Recherche en Informatique)
The defense will take place at 15h30 - Amphithéâtre 2 Bâtiment 2A IUT de Bordeaux - Campus Gradignan 35 rue de Naudet, 33175 GRADIGNAN
in front of the jury composed of
- Olivier LY - Professeur des universités - Université de Bordeaux - Directeur de these
- Nicolas MANSARD - Directeur de recherche - LAAS-CNRS - Rapporteur
- Sao Mai NGUYEN - Associate Professor - ENSTA - Rapporteur
- Jacky BALTES - Full professor - National Taiwan Normal University (NTNU) - Examinateur
Humanoid Robot Locomotion, Footstep Planning, and Fall Resilience via Reinforcement Learning Policies explores how to endow humanoid robots with robust locomotion and rapid fall recovery without relying on brittle heuristics or hand-crafted keyframes. The work addresses a central challenge in robotic autonomy: real humanoids must operate reliably in uncertain, contact-rich environments on limited onboard compute. Traditional model-based pipelines often lack adaptability, while deep reinforcement learning (DRL) offers the promise of data-driven, generalizable behaviors. The thesis thus investigates how to design DRL policies that are both computationally efficient and transferable zero-shot from simulation to physical robots, while integrating seamlessly into conventional locomotion stacks. Methodologically, the thesis develops foundational RL algorithms and robotics-oriented architectures, leading to two primary contributions trained in simulation with domain randomization and deployed on small humanoid robots. FootstepNet is an efficient actor-critic footstep planner that generates continuous, task-oriented step placements and, via its critic, predicts the number of steps required to reach multiple local goals---enabling rapid, upstream decision-making. It eliminates reliance on discrete footstep sets and fragile heuristics, supports onboard inference, and matches or surpasses ARA* baseline performance with substantially lower computational cost. FootstepNet was validated both in simulation and on hardware during RoboCup 2023. FRASA (Fall Recovery and Stand-up Agent) is a unified, end-to-end policy for fall recovery that directly maps proprioceptive inputs to motor commands, first establishing stabilizing ground contacts before transitioning to a standing pose. Leveraging the Cross-Q algorithm and exploiting robot symmetry, FRASA reduces training time to roughly 30 minutes and transfers zero-shot to real robots, outperforming a keyframe baseline while handling a wide variety of initial postures. Overall, the thesis demonstrates that lightweight, modular DRL policies can achieve practical, safe control for embedded humanoid systems, substantially reducing downtime after disturbances and paving the way for general, learning-based whole-body autonomy in real-world settings.
Contribution of deep learning to the numerical evaluation of dosimetry in electroporation
by Kylian DESIER (IMB - Institut de Mathématiques de Bordeaux)
The defense will take place at 13h30 - Salle de conférence 351 cours de la Libération Bâtiment A33 33400 Talence
in front of the jury composed of
- Baudouin DENIS DE SENNEVILLE - Directeur de recherche - Institut de Mathématiques de Bordeaux - Centre National de la Recherche Scientifique - Directeur de these
- Jing-Rebecca LI - Directrice de recherche - Unité de Mathématiques Appliquées (UMA) - Inria Saclay - Rapporteur
- Abdallah EL HAMIDI - Maître de conférences - Laboratoire des Sciences de l'Ingénieur pour l'Environnement (LaSIE) - La Rochelle Université - Rapporteur
- Damien VOYER - Maître de conférences - Laboratoire LERPA - EIGSI - Examinateur
- Lisl WEYNANS - Professeure des universités - Institut de Mathématiques de Bordeaux - Université de Bordeaux - Examinateur
- Clair POIGNARD - Directeur de recherche - Inria Rennes - CoDirecteur de these
Electroporation-based therapies represent a promising approach for the treatment of deep-seated tumors, particularly those of the liver, pancreas, and kidney. This technique relies on the application of high-intensity, short-duration electric pulses that modify the permeability of cell membranes. Depending on the parameters applied, these pulses can induce reversible electroporation, facilitating intracellular delivery of therapeutic molecules (electrochemotherapy), or irreversible electroporation (IRE), which causes targeted cell destruction without significant thermal effects. IRE is thus a non-thermal ablation technique capable of selectively eliminating tumor cells while preserving surrounding anatomical structures. For certain deep tumors, it may even constitute the only curative alternative where chemotherapy remains palliative. Despite its advantages, the clinical adoption of IRE remains limited due to challenges in precise treatment planning. The distribution of the electric field depends on numerous factors—physical parameters, tissue properties, and anatomical geometry—whose control determines both treatment efficacy and patient safety. Numerical modeling of the phenomenon, described by Maxwell's equations in the quasi-static regime, is therefore essential to predict the electric field and delineate ablation zones. However, the three-dimensional resolution of these equations on fine meshes remains computationally expensive. Classical numerical approaches, such as finite differences or finite elements, often conflict with clinical constraints, where simulations must be available within minutes on standard workstations. In this context, the present thesis—at the crossroads of applied mathematics, numerical modeling, and artificial intelligence—aims to design models combining physical accuracy, numerical precision, and computational efficiency to make IRE planning compatible with clinical use. The first part explores the potential of deep learning methods for solving partial differential equations (PDEs). A detailed review presents the main existing approaches, their theoretical foundations, strengths, limitations, and applicability to computational physics. The second part introduces a neural network model designed to solve the equation governing electric field propagation during an electroporation procedure. It describes the training protocol, datasets, evaluation metrics, and systematic comparison with reference numerical solutions. The third part, which constitutes the main contribution, develops an original hybrid model combining deep learning with an iterative numerical scheme. The neural network provides a fast initial estimation of the electric field, subsequently refined by a physical solver ensuring consistency with electrostatic laws. This approach combines the speed of the neural model with the accuracy of numerical computation. Finally, the fourth part presents the integration of this method into a complete software tool, tested on clinical data from real IRE cases. The results confirm the feasibility and robustness of the model, achieving accuracy comparable to standard methods while reducing computation time by about an order of magnitude. These performances enable dosimetry calculations in just a few tens of seconds on a conventional computer, paving the way for personalized and interactive therapeutic planning. In conclusion, this research demonstrates the potential of artificial intelligence, and particularly deep learning, to accelerate and enhance numerical simulation in interventional radiology. It heralds the emergence of a new generation of planning tools that are fast, accurate, and physically consistent, with the capacity to transform the clinical practice of electroporation-based therapies and, more broadly, image-guided interventional treatments.
ED Sciences Chimiques
Polyethylene deconstruction and functionalization through C-H activation.
by Lucile CLUZEAU (Institut des Sciences Moléculaires)
The defense will take place at 9h30 - Salle de conférence Bâtiment A12, Université de Bordeaux, 351 Cr de la Libération, 33405 Talence
in front of the jury composed of
- Yannick LANDAIS - Professeur des universités - Université de Bordeaux - Directeur de these
- Daniel TATON - Professeur des universités - Université de Bordeaux - CoDirecteur de these
- Stéphane GASTALDI - Directeur de recherche - CNRS - Examinateur
- Emmanuel BEYOU - Professeur des universités - Université Claude Bernard Lyon 1 - Rapporteur
- Isabelle DEZ - Professeure des universités - Université de Caen - Rapporteur
Polyethylene (PE), the most widely produced synthetic polymer worldwide, is renowned for its exceptional durability but remains notoriously resistant to chemical modification and upcycling due to the inert nature of its all–carbon backbone. In this work, we have developed a general, metal-free platform for the direct C–H functionalization of PE, featuring oxygen-tunable selectivity and affording either oxime- or ketone-functionalized polymers depending on the reaction conditions. This approach operates both (i) in solution, under visible-light irradiation in the presence of a photosensitizer, and (ii) in the melt state, within a twin-screw extruder under thermally activated conditions, thereby demonstrating its compatibility with both laboratory and industrial environments. The key reagent in this strategy is tert-butyl nitrite (t-BuONO), which, under visible-light irradiation, generates a transient alkoxy radical acting as a hydrogen-atom abstraction agent and a persistent nitrosyl radical responsible for the selective installation of oxime functionalities. Detailed mechanistic investigations, including quantum-yield determination and time-resolved transient absorption spectroscopy, have elucidated the energy-transfer activation mechanism of t-BuONO and identified the radical pathway involved in C–H activation. Furthermore, the introduced oxime groups are not terminal motifs but versatile intermediates that can be further transformed into esters, ethers, or urethanes, enabling fine control over the surface polarity and thermomechanical properties of the resulting polymer.
Engineering Chirality at Electrochemical Surfaces for Energy Conversion and Asymmetric Synthesis
by Zikkawas PASOM (Institut des Sciences Moléculaires)
The defense will take place at 14h00 - Salle E136 Vidyasirimedhi Institute of Science and Technology (VISTEC), 555 Moo 1 Payupnai, Wangchan District, Rayong, THAILAND 21210
in front of the jury composed of
- Alexander KUHN - Professor - Université de Bordeaux - Directeur de these
- Jeanne CRASSOUS - Directrice de recherche - CNRS Rennes - Rapporteur
- Hiroshi YAMAMOTO - Professor - Institute for Molecular Science - Rapporteur
- Supawadee NAMUANGRUK - Directrice de recherche - National Nanotechnology Center - Examinateur
- Chularat WATTANAKIT - Associate Professor - Vidyasirimedhi Institute of Science and Technology - CoDirecteur de these
Engineering electrochemical surfaces has emerged as a powerful concept for addressing critical challenges like clean energy and advanced chemical manufacturing. Simultaneously, the global demand for enantiopure chiral compounds. Addressing this need, our research introduces the design of a novel class of materials, namely chiral imprinted mesoporous metals. Recently we have successfully fabricated these materials with a diverse range of compositions. However, a significant challenge inherent to this approach has been the reliance on electrodeposition for material synthesis. The first part of this thesis focuses on developing a new, more versatile synthesis route for chiral encoded metals. Herein, we demonstrate the successful elaboration of these chiral imprinted materials using electroless deposition. In this part, homogeneous non-noble metal films containing chiral recognition sites combined with mesoporous features, were successfully generated on a nickel foam support. The electrodes demonstrated significant catalytic efficiency for the asymmetric electrosynthesis of chiral products, achieving a high enantiomeric excess (e.e.) of up to 85%. The promising results provide crucial foundational knowledge for future research, including optimizing deposition parameters to further enhance the reproducibility and long-term stability of these catalysts. The second part of this thesis extends the application of such designer materials to next-generation renewable energy technologies. Specifically, the Oxygen Reduction Reaction (ORR), a critical process in fuel cells. In this study, we leveraged the Chiral Induced Spin Selectivity (CISS) effect, which allows for spin-controlled catalysis. By pioneering a novel material-centered approach by applying chiral-encoded mesoporous platinum-iridium (Pt-Ir) alloys as electrocatalysts for ORR. These advanced materials were generated via an innovative electrodeposition. Electrochemical investigations demonstrate a significant enhancement in ORR activity on these intrinsically chiral surfaces, revealing significantly increased current intensities compared to both achiral analogs and electrodes imprinted with enantiomers of opposite chirality. This work introduces a new paradigm for designing highly efficient ORR catalysts. Finally, we illustrate an additional application of the CISS effect by introducing this concept to electrochemiluminescence (ECL), a powerful analytical technique used to generate light from electrochemical reactions. We propose a novel approach that integrates chiral mesoporous metal architectures with spin-selective properties to enhance ECL sensitivity. This chapter introduces a spin-selective ECL approach by integrating the spin-filtering effect of chiral-imprinted nickel electrodes on the ORR with the ECL emission of L-012, a luminol derivative that emits blue light via a cathodic mechanism. We demonstrate that imprinting the electrode with a specific enantiomer promotes the ECL emission. Furthermore, we explored the exciting realm of wireless electrochemistry by integrating our chiral imprinted electrodes into a bipolar electrochemical setup. This work represents the first investigation of CISS-enhanced materials for the specific purpose of amplifying ECL signals, opening new avenues for innovative advancements in the field. In summary, this thesis presents a holistic approach to engineering chiral electrochemical surfaces by developing novel synthesis methods and demonstrating their utility. We have successfully enhanced both asymmetric synthesis and key energy conversion processes. The application of the CISS effect has been shown to boost the efficiency of both the ORR and ECL. The findings presented here provide fundamental insights and a strong foundation for future research in advanced materials, catalysis, and spin-controlled electrochemistry.
Topotactic fluorination of equiatomic ternary silicides: synthesis, structural characteristics and electronic properties
by Thanh Hung Thinh TRUONG (ICMCB - Institut de Chimie de la Matière Condensée de Bordeaux)
The defense will take place at 14h00 - Amphithéâtre ICMCB, 87 Avenue du Dr Albert Schweitzer, 33600 Pessac
in front of the jury composed of
- Sophie TENCE - Chargée de recherche - Institut de Chimie de la Matière Condensée de Bordeaux - Directeur de these
- Pierric LEMOINE - Professeur des universités - Université de Lorraine - Rapporteur
- Katia GUERIN ARAUJO DA SILVA - Professeure des universités - Institut de chimie de Clermont Ferrand - Rapporteur
- Jérôme LHOSTE - Maître de conférences - Institut des Molécules et Matériaux du Mans - Examinateur
- Cyril AYMONIER - Directeur de recherche - Institut de Chimie de la Matière Condensée de Bordeaux - Examinateur
This PhD work focused on the investigation of topotactic fluorine insertion into ternary 111-type intermetallic compounds, specifically those adopting two layered structural types: CeFeSi-type (P4/nmm) and CeScSi-type (I4/mmm). For the CeFeSi-type family, fluorination experiments were carried out on LaRuSi and CeRuSi using dynamic synthesis under a continuous flow of C4F8. Under almost all investigated conditions, fluorine incorporation was nearly identical for both systems, around 0.5 for La- and 0.6 for Ce-based compounds, leading to a superstructure with a doubled c parameter. A slight distortion of the ab basal plane, leading to a symmetry reduction from P4/nmm (tetragonal) to Pmmn (slightly orthorhombic), results in an ordered arrangement of fluorine ions at alternating tetrahedral sites (occupied and vacant) within the same ionic layer. This induces significant modifications in the [RuSi]p- blocks, thereby stabilizing two distinct RuSi4 tetrahedral environments. In Ce-based compounds, stacking faults associated with variable fluorine occupancies at tetrahedral sites were detected by HRTEM and modeled using the FAULTS program. Magnetically, LaRuSiFx exhibits a Curie-Weiss-type behavior at high temperatures together with a metallic behavior, in contrast to the Pauli paramagnetism observed for the pristine LaRuSi. The CeRuSiFx phases display a similar high-temperature magnetic response; however, at low temperatures, a Schottky anomaly (Ce/La) emerges, competing with magnetic ordering (Ce). In the second part, the topotactic fluorination was extended to LaScSi (CeScSi-type structure), which contains two distinct interstitial environments: tetrahedral La4 and octahedral Sc4La2 vacant sites. Three synthetic strategies were employed: dynamic fluorination under a continuous C4F8 flow, static fluorination under a C4F8 atmosphere, and a new solid–solid fluorination method using a (VDF–TFE)-based oligomer. In all cases, fluoride ions were incorporated exclusively into the La4 tetrahedral sites. For the C4F8-based routes, co-insertion of other light elements such as H and O was systematically observed, due to water dissociation during the reaction, resulting in relatively low F contents (≈ 0.09–0.14) in all annealed samples. Notably, under static conditions, a metastable high-F phase (F ≈ 0.42) was detected; however, it decomposed upon annealing, releasing fluorine and stabilizing anionic vacancies at the tetrahedral sites. Finally, a novel solid–solid fluorination pathway was developed using a (VDF–TFE)-based oligomer. This approach enabled full occupancy of the tetrahedral site by fluorine, yielding a previously unprecedented hydride–fluoride phase, LaScSiH0.5F1.0, characterized by an exceptionally expanded c parameter.
ED Droit
Criminal social values
by Julie VAURET (INSTITUT DE SCIENCES CRIMINELLES ET DE LA JUSTICE)
The defense will take place at 14h00 - 1K Pôle juridique et judiciaire de l'université de Bordeaux 35, place Pey-Berland 33000 Bordeaux
in front of the jury composed of
- Jean-Christophe SAINT-PAU - Professeur agrégé - Université de Bordeaux - Directeur de these
- François ROUSSEAU - Professeur agrégé - Université de Nantes - Rapporteur
- Xavier PIN - Professeur agrégé - Université Jean Moulin Lyon 3 - Rapporteur
- Agathe LEPAGE - Professeure agrégée - Université Paris-Panthéon-Assas - Examinateur
- Maxime BRENAUT - Professeur agrégé - Université de Bordeaux - Examinateur
- Bertrand DE LAMY - Professeur agrégé - Université de Toulouse - Examinateur
Although their protection can be seen as the aim of criminal law and, as such, as its core, criminal social values are ofter mentioned but rarely studied and their content defined. It is true that they have been criticized et analyzed when defined by their place among the elements of offenses, however their content is still very much silenced. As part of a «criminal public order», criminal social values' analysis must be turned towards their identification and the determination of the relationships they maintain amongst themselves.
ED Entreprise Economie Société
INFLUENCE OF SOCIODEMOGRAPHIC FACTORS IN THE EVALUATION OF WINE ATTRIBUTES IN A COLLECTIVIST CULTURAL CONTEXT: The case of Bordeaux wine in Cameroon, Ivory Coast, Nigeria, and Kenya
by Mathieu NLEND MOMNOUGUI (IRGO - Institut de Recherche en Gestion des Organisations)
The defense will take place at 9h30 - c 107 IAE - Ecole Universitaire de Management IRGO 35 avenue Abadie - CS 51412 Bat C 33072 Bordeaux Cedex
in front of the jury composed of
- Franck DUQUESNOIS - Maître de conférences - Université de Bordeaux - Directeur de these
- Jean-Philippe GALAN - Professeur des universités - Université de Bordeaux IAE - Ecole Universitaire de Management - CoDirecteur de these
- Mbaye Fall DIALLO - Professeur des universités - IAE - Université de Lille - Rapporteur
- Foued CHERIET - Professeur des universités - Institut Agro Montpellier - Rapporteur
- Franck CELHAY - Professeur - Montpellier business school - Examinateur
- Catherine VIOT - Professeure des universités - IAE - Université de Bordeaux - Examinateur
For several decades, international marketing has been confronted with the issue of culture's influence on consumer perception. A large amount of research in this field has focused on this topic, with Hofstede's theory of national cultural dimensions being the most widely used. This study analyzes the influence of sociodemographic factors on consumer sensitivity to the attributes of Bordeaux wine in an African context (Cameroon, Ivory Coast, Kenya, Nigeria) dominated by collectivism, high hierarchical distance, and short-term orientation. A Best Worst Scale approach was used for this research. The results from the analysis of data from an experiment conducted on 813 subjects indicate that the structuring of preferences in the African context is mainly organized around the reduction of perceived risk through relational and experiential signals. Price has a powerful but contextual effect, indexed to the purchasing occasion (socializing with friends versus a more formal event), while the more technical aspects of the product have a greater influence mainly among more engaged and higher-income segments.
ED Sciences de la Vie et de la Santé
Deciphering the metabolic fingerprint of grapevine berries under high temperature
by Xi ZHAN (Ecophysiologie et Génomique Fonctionnelle de la Vigne)
The defense will take place at 14h00 - Amphithéâtre de l'ISVV 210 Chemin de Leysotte, 33140 Villenave-d'Ornon
in front of the jury composed of
- David LECOURIEUX - Maître de conférences - Université de Bordeaux - Directeur de these
- Marielle ADRIAN - Professeure - Université de Bourgogne Europe - Rapporteur
- Nathalie GAVEAU - Professeure - Université de Reims Champagne-Ardenne - Rapporteur
- Bertrand GAKIèRE - Maître de conférences - Université de Paris-Saclay - Examinateur
- Kentaro MORI - Maître de conférences - Université de Bordeaux - Examinateur
- Marie-France CORIO-COSTET - Directrice de recherche - INRAE Nouvelle-Aquitaine Bordeaux - Examinateur
This work, conducted within the UMR EGFV (QualyGrapE team), aims to elucidate the metabolic adjustments that enable berries to respond to high temperatures (HT). Initially focused on berry redox balance under HT, the study was subsequently reoriented toward a broader metabolic characterization due to both technical limitations in berry enzymatic assays and the inherent complexity of functional characterization of genes in grapevine. These experimental constraints, however, opened a more integrative perspective for my PhD project: since metabolites act as the bridge between genotype and phenotype, they not only determine grape quality attributes such as color, flavor, and aroma, but also play direct roles in stress perception, signaling, and defense. Consequently, metabolite profiling emerged as a powerful and biologically meaningful approach to investigate how grape berries remodel their metabolism under HT. The research comprises three main parts. The first compares Cabernet-Sauvignon and Merlot, two cultivars differing in their putative thermosensitivity, with a focus on polyphenolic compositions. Using targeted LC-MS analyses at distinct developmental stages under controlled conditions, we established a detailed polyphenolic profile of both cultivars under varying HT durations. The second part expands this comparison through untargeted LC-MS and targeted GC-MS metabolomic approaches, capturing a broad range of primary and secondary metabolites and enabling the identification of robust metabolite markers associated with heat exposure. Finally, the study tentatively explores the functional characterization of a peroxidase gene potentially involved in redox regulation and anthocyanin degradation under HT. A CRISPR-Cas9-based strategy was developed to generate stable transgenic microvine lines for functional validation. Together, our findings provide new insights into the metabolic remodeling of grape berries under high temperature and highlight both common and cultivar-dependent responses. The identified metabolite markers offer valuable leads for developing viticultural strategies and breeding programs aimed at improving heat resilience in grapevine.
ED Sciences Physiques et de l'Ingénieur
Study of the equation of state of boron and boron nitride at very high pressure by shock waves produced by laser
by Diluka SINGAPPULI HEWAGE (Centre Lasers Intenses et Applications)
The defense will take place at 10h00 - Amphithéâtre G Campus Peixotto Bâtiment A29 RDC 351 Cr de la Libération, 33400 Talence
in front of the jury composed of
- Dimitri BATANI - Professeur des universités - Université de Bordeaux - Directeur de these
- Alessandra BENUZZI-MOUNAIX - Directrice de recherche - LULI, École Polytechnique, CNRS, CEA, UPMC - Rapporteur
- Michel BOUSTIE - Directeur de recherche - CNRS Institut Pprime ISAE ENSMA, Futuroscope - Rapporteur
- Norimasa OZAKI - Associate Professor - Graduate School of Engineering, Osaka University - Examinateur
- Stéphanie BRYGOO - Ingénieure chercheur - Commissariat a l'Energie Atomique - Examinateur
- Guillaume BOUTOUX - Ingénieur chercheur - CEA, DAM, DIF - Examinateur
- Emmanuel D'HUMIèRES - Professeur - Université de Bordeaux, CNRS, CEA, CELIA (Centre Lasers Intenses et Applications), UMR 5107 - Examinateur
Boron and its compounds are of major scientific interest in both astrophysics and inertial confinement fusion (ICF). In astrophysics, boron abundance provides a powerful probe to test models of cosmological nucleosynthesis, while materials such as boron carbide (B4C) and boron nitride (BN) are considered as alternative ablators to improve the efficiency of ICF. A precise knowledge of their equations of state (EOS) is essential to predict material behavior under extreme conditions. However, experimental data of EOS beyond the megabar regime (> 100 GPa) remain very limited, preventing a robust validation of theoretical models. This research falls within the broader field of Warm Dense Matter (WDM), an intermediate regime between condensed matter and plasma, characterized by near-solid densities and temperatures of a few electronvolts. This domain, which is crucial for understanding planetary interiors and fusion plasmas, remains particularly challenging to model, emphasizing the need for new experimental benchmarks. In this context, this thesis focused on the EOS of boron nitride in hexagonal and cubic phases at very high pressures. Experiments were carried out on large-scale laser facilities in Europe (PALS, ELI-Beamlines) and worldwide (SGIII-P, GEKKO XII), relying on optical diagnostics such as VISAR and SOP. The experimental results were compared with ab initio simulations (DFT-MD), semi-empirical models (FEOS), and hydrodynamic calculations. In parallel, dynamic compression experiments of pre-compressed water were conducted, allowing us to approach its cold curve, establish a direct connection with planetary interior modeling, and provide a new route to probe the WDM regime. These experimental studies enrich the EOS database of boron nitride and dense water, and provide essential constraints for the theoretical modeling of Warm Dense Matter, with direct implications for astrophysics and the development of inertial confinement fusion.
ED Sciences et environnements
Satellite-derived shoreline data assimilation for sandy coast evolution
by Georgios AZORAKOS (Environnements et Paléoenvironnements Océaniques et Continentaux)
The defense will take place at 14h00 - PABA RDC bâtiment B5, 33600 Pessac
in front of the jury composed of
- Bruno CASTELLE - Directeur de recherche - Université de Bordeaux - Directeur de these
- Marissa YATES - Chargée de recherche - Ecole des Ponts / Laboratoire d'Hydraulique Saint-Venant - Rapporteur
- Rafael ALMAR - Directeur de recherche - Ird - Rapporteur
- Francesca RIBAS PRATS - Associate Professor - Universitat Politècnica de Catalunya - Examinateur
- Stéphane BERTIN - Chargé de recherche - CNRS - Examinateur
- Didier SWINGEDOUW - Directeur de recherche - UMR CNRS 5805 EPOC - OASU - Université de Bordeaux - Examinateur
- Déborah DIDIER - Directrice de recherche - BRGM - CoDirecteur de these
This thesis addresses the critical challenge of modelling and predicting the future evolution of sandy shorelines. Traditional shoreline change models require extensive in-situ data for calibration and are therefore limited to a few well-monitored sites worldwide. To overcome this limitation, this work explores the use of optical satellite imagery as a widely available source of shoreline observation to calibrate equilibrium shoreline models at cross-shore transport dominated sites. Focusing on Truc Vert beach in southwest France, where previous studies have demonstrated good model skill across time scales from individual storms to decades, an 11-year dataset (2009–2020) of satellite-derived waterlines was processed into satellite-derived shorelines (SDS) with varying water level corrections and alongshore averaging lengths. The results showed that SDS data can calibrate shoreline models with skill comparable to traditional in-situ shoreline data, and that alongshore averaging is the only essential data processing step, while other site-specific corrections provide minimal benefit. Building on this, a dual parameter/state ensemble Kalman filter (enKF) was implemented to track non-stationarity in model free parameters at three sites along the west coast of Europe, revealing that while introducing non-stationary parameters did not consistently outperform well-calibrated stationary models, the enKF correction steps substantially improved model skill, highlighting its potential for real-time data assimilation. The findings suggest that parameter variability may reflect unresolved processes and that climate-driven time-varying parameters could enhance extreme event prediction at seasonally dominated sites and improve performance in complex multimodal wave climates. The thesis culminates in the development of a new data-assimilation framework within the LX-Shore model, combining simulated annealing-based calibration with the assimilation of uncertain satellite-derived shoreline data. This framework enables the incorporation of shoreline boundary conditions at inlets and estuary mouths, identification of primary sources of model error, and improvement of shoreline predictions while accounting for data uncertainties. Overall, this work demonstrates the feasibility and potential of integrating remote sensing and data assimilation to achieve robust and scalable shoreline change modelling at data-limited coasts worldwide.
ED Sociétés, Politique, Santé Publique
Validity and relevance of blood biomarkers for Alzheimer's disease in a national clinical cohort
by Vincent BOUTELOUP (Bordeaux Population Health Research Center)
The defense will take place at 14h00 - Amphi Louis ISPED, Université de Bordeaux 146, rue Léo Saignat CS61292 33076 Bordeaux cedex - FRANCE
in front of the jury composed of
- Carol BRAYNE - Professeure des universités - Université de Cambridge - Rapporteur
- Bernard HANSEEUW - Professeur des universités - praticien hospitalier - Université catholique de Louvain - Rapporteur
- Aurélie BEDEL - Professeure des universités - praticienne hospitalière - CHU Bordeaux - Examinateur
- Rodolphe THIEBAUT - Professeur des universités - praticien hospitalier - Université de Bordeaux - Examinateur
Alzheimer's disease (AD), responsible for 60 to 70% of dementia cases, is a multifactorial neurodegenerative disorder characterized by amyloid plaque deposits and neurofibrillary degeneration. With 55 million people living with dementia worldwide in 2020 (1.4 million in France) and projections reaching 150 million by 2050, the search for accessible and reliable biomarkers is a major challenge to optimize diagnostic strategies—especially in anticipation of effective early treatments. Blood biomarkers, such as phosphorylated tau proteins (p-tau181, p-tau217) and the Aβ42/Aβ40 ratio, could provide an alternative to current reference tests (lumbar puncture, PET imaging). However, their interpretation remains complex due to their low concentration and sensitivity to variability factors (e.g., age, renal function). This thesis, based on data from the MEMENTO cohort (2,323 non-demented participants followed for five years in 26 Memory Centers), explores the validity and clinical relevance of AD blood biomarkers. A first analysis of the factors associated with blood biomarker concentrations revealed that their variability was primarily explained by age, and to a lesser extent by clinical markers of AD (cognition, brain atrophy, genetic). These findings suggested that interpreting these biomarkers in relation to AD progression should not be done in isolation but in conjunction with the patient's overall clinical profile. A second study focused on interpreting p-tau217 concentrations based on patients' clinical phenotypes, highlighting higher positive predictive values for common AD phenotypes compared to others. The final study employed various methodological approaches to address the issue of changing biomarker assay methods over time in a cohort designed to examine their repeated measurements. We demonstrated the value of latent process models, which do not require dedicated biological material, unlike transferability ("bridging") studies. Integrating blood biomarkers into clinical practice offers numerous advantages but requires real-world prospective studies to assess their clinical utility, impact on patient management, and cost-effectiveness. The MEMENTO cohort, with its rigorous methodology and clinical diversity, provides an ideal framework for studying these biomarkers. However, its limitations (participant selection, availability of reference tests (PET and CSF) in subsamples) underscore the need to validate results in more representative populations. In conclusion, blood biomarkers—particularly p-tau217—represent a major advance in AD mechanism research. Their use in memory clinics could be implemented if it accounts for comorbidities and the detailed clinical presentation of patients.
Reasoning with imperfect and temporal clinical data: logic-based approaches for phenotyping and medical event inference
by Kokou AWUKLU (Bordeaux Population Health Research Center)
The defense will take place at 14h00 - Amphithéâtre du LaBRI Domaine universitaire, 351 cours de la Libération, 33405 Talence
in front of the jury composed of
- Fleur MOUGIN - Professeure des universités - Université de Bordeaux - CoDirecteur de these
- Stefan SCHULZ - Full professor - Université de Graz - Rapporteur
- Adrien COULET - Chargé de recherche - Inria Paris - Rapporteur
- Meghyn BIENVENU - Directrice de recherche - Université de Bordeaux - CoDirecteur de these
- Maria Vanina MARTINEZ - Chargée de recherche - Université autonome de Barcelone - Examinateur
- Michael SIOUTIS - Professeur - Université de Montpellier - Examinateur
- Pascal STACCINI - Professeur des universités - praticien hospitalier - Université Côte d'Azur - Examinateur
The adoption of electronic health records has considerably enhanced access to large volumes of clinical data. While this accessibility is invaluable for both healthcare delivery and research, it also introduces new challenges arising from the complexity of medical data. These challenges include its implicitness (i.e., the need for domain expertise to interpret data), imperfections such as inconsistency, uncertainty, and incompleteness, and its inherently temporal nature. This thesis investigates how logic-based approaches can address these challenges. First, I investigated an ontology-driven approach to illustrate how ontologies can be used to evaluate medical data quality, with a focus on lung cancer phenotyping. This involved designing an ontology to capture essential domain knowledge and apply- ing it to query the Clinical Data Warehouse of Bordeaux University Hospital. The work highlighted both the benefits of ontologies in representing domain knowledge and identifying inconsistencies, as well as their limitations, particularly in handling temporally inconsistent healthcare data. Building on this experience, I then proposed a novel logic-based framework for inferring high-level events from temporal clinical data, in a way that better aligns with clinical reasoning and decision-making. The framework defines logical rules speci- fying the existence conditions of an event at a given time-point, along with optional termination conditions that signal its possible end. It also introduces two aggregation methods to construct event intervals from these conditions. Furthermore, the formal- ism supports the definition of meta-events, obtained by combining or generalizing other events, and integrates confidence levels and a repair mechanism to handle imperfec- tions in event detection. To validate the framework, I implemented its core components using Answer Set Programming, a declarative logic programming paradigm, and evalu- ated the resulting system, CASPER, on two medical use cases. The evaluation showed both computational feasibility and alignment with expert medical opinions.