HAL Portal Univ Catholille
Not a member yet
19763 research outputs found
Sort by
Détection d'images générées par l'IA à l'aide de l'apprentissage profond multimodal
No full textGenerative artificial intelligence (AI) and its applications offer significant benefits but raise potentially critical societal and ethical concerns. This thesis explores the field of synthetic content generation and detection, with a particular focus on synthetic images. These are AI-generated images that can appear indistinguishable from real photographs, creating new challenges for digital forensics, media integrity, and public trust. While generative models such as generative adversarial networks (GANs) and diffusion models have evolved rapidly, most existing detection methods remain limited in terms of generalizability, robustness, and interpretability. To address these challenges, this research investigates new AI-driven frameworks to improve the generalizability, robustness, and interpretability of current image detection methods.The various aspects of synthetic image detection are covered in this thesis by four key contributions. The first contribution, Bi-LORA, presents an efficient vision-language approach that reformulates the detection problem into an image caption task, achieving strong zero-shot generalization across unseen generative models. The second, RAVID, proposes a retrieval-augmented visual detection framework that enhances robustness and interpretability by incorporating external visual context. The third, DeeCLIP, presents a lightweight transformer-based model that fuses shallow and deep features to improve resistance to image degradation and post-processing. Finally, FIDAVL unifies synthetic image detection and source attribution in a single multitasking framework based on the soft prompt tuning of vision-language models.This thesis provides an in-depth analysis of the current state of synthetic image research and its societal impacts. It emphasizes the urgent need for effective and robust detection methods. The approaches introduced advance the scientific community and could benefit practical applications in media authenticity and security. This research marks an important step forward in developing solutions to the ethical challenges posed by GenAI.L'intelligence artificielle générative (IA) et ses applications offrent des avantages considérables, mais soulèvent des questions sociétales et éthiques potentiellement critiques. Cette thèse explore le domaine de la génération et de la détection de contenus synthétiques, en mettant particulièrement l'accent sur les images synthétiques. Il s'agit d'images générées par l'IA qui peuvent paraître indiscernables de photographies réelles, ce qui crée de nouveaux défis pour la criminalistique numérique, l'intégrité des médias et la confiance du public. Alors que les modèles génératifs tels que les réseaux antagonistes génératifs (GANs) et les modèles de diffusion ont connu une évolution rapide, la plupart des méthodes de détection existantes restent limitées en termes de généralisation, de robustesse et d'interprétabilité. Pour relever ces défis, cette recherche étudie de nouvelles approches basées sur l'IA afin d'améliorer la généralisation, la robustesse et l'interprétabilité des méthodes actuelles de détection d'images.Les différents aspects de la détection d’images synthétiques sont examinés dans cette thèse à travers quatre contributions majeures. La première contribution, intitulée Bi-LORA, propose une approche vision–langage efficace qui reformule le problème de détection sous la forme d’une tâche de génération de légendes d’images. Cette méthode démontre une capacité de généralisation zero-shot remarquable face à des modèles génératifs encore non vus. La deuxième contribution, nommée RAVID, introduit un cadre de détection visuelle enrichi par la recherche d’images (retrieval-augmented visual detection). Ce cadre vise à renforcer la robustesse et l’interprétabilité des systèmes de détection en intégrant un contexte visuel externe pertinent. La troisième contribution, DeeCLIP, présente un modèle léger fondé sur une architecture de type transformeur, qui combine des caractéristiques superficielles et profondes afin d’améliorer la résilience aux dégradations visuelles et aux opérations de post-traitement. Enfin, la quatrième contribution, FIDAVL, propose une approche unifiée de la détection d’images synthétiques et de l’attribution à la source, au sein d’un cadre multitâche reposant sur l’ajustement souple des invites (soft prompt tuning) dans les modèles vision–langage.Cette thèse propose une analyse approfondie de l’état actuel de la recherche sur les images synthétiques ainsi que de leurs impacts sociétaux. Elle souligne l’importance et l’urgence de concevoir des méthodes de détection à la fois efficaces et robustes. Les approches développées dans ce travail contribuent à l’avancement de la recherche scientifique et présentent un potentiel d’application concret dans les domaines de l’authenticité des médias et de la sécurité numérique. Dans cette perspective, cette recherche constitue une étape significative vers le développement de solutions face aux défis éthiques soulevés par le GenAI
Identification of antihypertensive, antidiabetic, and antioxidant peptides derived from hydrolysates of dairy white wastewaters containing milk proteins using machine learning insights
No full textInternational audienceDairy white wastewater (WW), a by-product of industrial cleaning processes, contains residual milk proteins that can be enzymatically converted into bioactive peptides. In this study, WW proteins were hydrolyzed using four enzymes, pepsin, trypsin, thermolysin, and pronase E, for up to 240 min, and the resulting hydrolysates were evaluated for ACE inhibition, DPP-IV inhibition, and antioxidant capacity. Thermolysin hydrolysates exhibited the strongest ACE inhibition, with IC50 values as low as 21.0 μg protein/mL, whereas pepsin and pronase E hydrolysates showed DPP-IV inhibitory activities with IC50 values of 2.4–3.1 mg protein/mL. Pepsin hydrolysates presented the highest antioxidant capacity, reaching 3.5 μM Trolox equivalents/mg protein. LC-MS/MS analysis combined with multivariate statistics identified 60 discriminant peptides, including 17 peptides previously reported to possess antihypertensive, antidiabetic, and/or antioxidant bioactivities. Based on a combination of PLS-DA loadings, QSAR scores, novelty relative to known bioactive peptides and physicochemical diversity, 20 peptides were synthesized and validated experimentally. Several peptides such as LRF, QW, GAWY, PPF, GPIVL, and SFNPTQL exhibited potent inhibitory effects, with micromolar IC50 values for ACE and/or DPP-IV. In comparison to chemically synthesized ACE inhibitors like captopril, the most potent peptide, LRF, is five times more active on a molar basis (11.34 μM vs 63.06 μM). These findings demonstrate that WW is a promising source of multifunctional peptides and that integrating peptidomics with machine learning accelerates peptide discovery and validation
Comment on “Influence of layer thickness on time domain Brillouin scattering oscillation amplitude in multilayer films” [J. Appl. Phys. 136, 225302 (2024)]
No full textInternational audiencePicosecond acoustics allows for the probing of multilayered nanometric structures, with applications in various fields of fundamental research. In a recent study, Zhang et al.1 have reexamined the optical cavity effect that impacts time-domain Brillouin acoustic spectroscopy (TDBS) in such systems. This interference phenomenon arises from Fabry–Pérot cavities formed by the parallel interfaces within the stack. This has long been modeled either through a general solution based on Green’s functions or via analytical expressions derived from standard Fresnel formulas. In their study, Zhang et al. examined the Brillouin signal in a series of amorphous SiO2 layers grown on a Si wafer and capped with a metal Al transducer. This configuration had been thoroughly investigated as part of a larger effort to study high-frequency acoustic losses in silica glass by Ayrinhac et al. in Ref. 2. However, when attempting to reproduce these results, Zhang et al. failed to achieve quantitative agreement with the calculated expectations, due to poorly characterized samples and likely inadequate data processing. We show that simple optical reflectance measurements combined with appropriate data normalization can, in fact, fully account for such observations
Unraveling how mycorrhizal inoculation shapes the wheat foliar transcriptome and enhances resistance to septoria tritici blotch
No full textInternational audienceArbuscular mycorrhizal fungi (AMF), known to form mutualistic associations with most terrestrial plant species, could represent promising ecological alternatives to agrochemicals, which are harmful to both human health and environment. Although most studies investigate the effectiveness of mycorrhizal symbiosis in controlling root diseases, its influence on foliar diseases is still poorly explored, particularly in wheat. A prior investigation conducted under controlled conditions revealed that inoculation with the AMF species Funneliformis mosseae (Fm) protects wheat leaves from both pathogenic fungi, the biotrophic Blumeria graminis f. sp. tritici, responsible for powdery mildew (Mustafa et al, 2016, 2017), and the hemibiotrophic Zymoseptoria tritici (Zt), causing septoria tritici blotch (STB). This protection seems to be linked to the activation of "Mycorrhiza-Induced Resistance" (MIR), facilitating the systemic induction of plant defenses mediated by AMF (Allario et al., 2025). Nonetheless, the metabolic alterations that contribute to this resistance are poorly understood. In this study, we performed transcriptomic analyses to investigate how mycorrhizal symbiosis impacts gene expression in wheat leaves, resulting in plant protection against STB. Leaves from 6-week-old seedlings, inoculated or not with Fm and/or Zt, were sampled to study gene expression, 48 hours after Zt spores inoculation. Transcriptomic data showed a systemic induction of numerous genes in leaves of mycorrhizal plants under non-infectious conditions, many of which were common to those induced by Zt inoculation in non-mycorrhizal plants. These genes were involved in proteolysis, antioxidant mechanisms and plant immune response, particularly linked to jasmonic acid and ethylene signalling pathways. This induction of specific defense mechanisms by Fm before Zt inoculation (cell wall thickening, stomatal closure, defense proteins, maintenance of homeostasis and cellular function) was later completed by a downregulation of genes participating in photosynthesis and chloroplastic functions in mycorrhizal plants under infectious conditions. Together, these findings highlight AMF potential to induce MIR in wheat through systemic and coordinated metabolic reprogramming, leading to STB symptom reduction.Keywords : Wheat, septoria tritici blotch, arbuscular mycorrhizal fungi, Mycorrhiza-Induced Resistance, transcriptomic analysis
Toward Zero-Waste Valorization of Isatis tinctoria: A Preliminary Study Regarding Antioxidant Properties of Organic Extract
No full textInternational audienceIntroduction: Isatis tinctoria is a non-food Brassica plant that was extensively cultivated in Europe between the 12th and 17th centuries for the production of the blue indigo naturalis dye pigment. Presently, a project is underway to reintroduce the species to several European regions. However, the mere production of the pigment does not necessarily guarantee the economic viability of its cultivation. A considerable body of research has been dedicated to examining the antioxidant capacity of this plant. Extensive studies have demonstrated that extracts obtained from the cauline leaves exhibit notable antioxidant properties. Consequently, a direct incompatibility exists in the valorization of the plant between the production of pigment and the production of antioxidant compounds. Method: This study aims to circumvent this competitive dynamic; a zero-waste valorization of Isatis tinctoria can be posited. The antioxidant potential of both leaf residues after pigment extraction and roots was evaluated from a hydroponic culture of Isatis tinctoria. The antioxidant capacity of the obtained extracts was evaluated according to four colorimetric tests: ABTS, DPPH, FRAP, and CUPRAC. Results: The results of the study indicate that the re-use of the cauline leaves of Isatis tinctoria is possible, in particular through its DPPH (544 μmol TE/100g DW) and CUPRAC (807 μmol TE/100g DW) activities. Nevertheless, the drying process following pigment extraction must be enhanced to ensure the reproducibility of results. It has been observed that the roots of Isatis tinctoria exhibit a degree of interest, albeit to a more moderate extent, in terms of metal reduction activities (FRAP: 120 μmol TE/100g FW; CUPRAC: 250 μmol TE/100g FW). Discussion: However, it is imperative to optimize both the drying step and the extraction methodology. Furthermore, the components constituting both extracts must be characterized. Conclusion: The subsequent study demonstrated that Isatis tinctoria leaves, a byproduct of indigo extraction, can be valorized through a biocascade approach for the production of an antioxidant extract. Furthermore, the study suggests that plant zero-waste valorization can be achieved through the extraction of antioxidants from Isatis tinctoria roots
TIGRÉ : LA GUERRE OUBLIÉE DE LA CORNE DE L'AFRIQUE: Entre centralisation autoritaire et fédéralisme ethnique, un conflit de haute intensité aux ramifications régionales (2020-2022) Entre centralisation autoritaire et fédéralisme ethnique, un conflit de haute intensité aux ramifications régionales (2020-2022)
No full textInternational audienc
PCB-Based Hybrid Series/Corporate-Fed 4 × 4 D-Band Phased Array With Wide-Angle Scanning
No full textInternational audienceSixth generation (6G) wireless networks are envisioned to include aspects of energy footprint reduction (sustainability), besides those of network capacity and connectivity, at the design stage. This paradigm change requires radically new physical layer technologies. Notably, the integration of large-aperture arrays and the transmission over high frequency bands, such as the sub-terahertz spectrum, are two promising options. In many communication scenarios of practical interest, the use of large antenna arrays in the sub-terahertz frequency range often results in short-range transmission distances that are characterized by line-of-sight channels, in which pairs of transmitters and receivers are located in the (radiating) near field of one another. These features make the traditional designs, based on the far-field approximation, for multiple-input multiple-output (MIMO) systems sub-optimal in terms of spatial multiplexing gains. To overcome these limitations, new designs for MIMO systems are required, which account for the spherical wavefront that characterizes the electromagnetic waves in the near field, in order to ensure the highest spatial multiplexing gain without increasing the power expenditure. In this paper, we introduce an analytical framework for optimizing the deployment of antenna arrays in line-of-sight channels, which can be applied to paraxial and non-paraxial network deployments. In the paraxial setting, we devise a simpler analytical framework, which, compared to those available in the literature, provides explicit information about the impact of key design parameters. In the non-paraxial setting, we introduce a novel analytical framework that allows us to identify a set of sufficient conditions to be fulfilled for achieving the highest spatial multiplexing gain. The proposed designs are validated with numerical simulations
