Portail HAL Nantes Université
Not a member yet
120169 research outputs found
Sort by
Le littoral face à la montée de la mer : résister ou se replier? Quels futurs pour les littoraux habités ?
No full textInternational audienc
Enhancing breast cancer screening: Unveiling explainable cross-view contributions in dual-view mammography with Sparse Bipartite Graphs Attention Networks
No full textInternational audienceMedical imaging techniques like mammography enable early breast cancer detection and are part of regular screening programs. Typically, a mammogram exam involves two views of each breast, providing complementary information, but physicians rate the breast as a whole. Computer-Aided Diagnostic tools focus on detecting lesions in a single view, which is challenging due to high image resolution and varying scales of abnormalities. The projective nature of the two views and different acquisition protocols add complexity to dual-view analysis. To address these challenges, we propose a Graph Neural Network approach that models image information at multiple scales and the complementarity of the two views. To this end, we rely on a superpixel decomposition, assigning hierarchical features to superpixels, designing a dual-view graph to share information, and introducing a modified Sparse Graph Attention Layer to keep relevant dual-view relations. This improves interpretability of decisions and avoids the need to register pairs of views under strong deformations. Our model is trained with a fully supervised approach and evaluated on public and private datasets. Experiments demonstrate state-of-the-art classification and detection performance on Full Field Digital Mammographies, achieving a breast-wise AUC of 0.96 for the INbreast dataset, a sensitivity of 0.97 with few false positives per image (0.33), and a case-wise AUC of 0.92 for the VinDr dataset. This study presents a Sparse Graph Attention method for dual-view mammography analysis, generating meaningful explanations that radiologists can interpret. Extensive evaluation shows the relevance of our approach in breast cancer detection and classification
Shaping of the selectivity profile of single-scan ultraselective NMR experiments
No full textInternational audienc
Ultrafast Correlation Spectroscopy with multi-solvent suppression for the analysis of wine
No full textInternational audienc
Vers une édition sémantique de vidéos volumétriques: segmentation 4D basée Gaussian Splatting et moyens d’évaluation
No full textVolumetric video enables dynamic 3D representations of real-world scenes, supporting free-viewpoint exploration and natural 6 Degrees of Freedom (6 DoF) navigation. To achieve truly immersive interaction, semantic segmentation is essential for object-level understanding. While recent novel view synthesis methods, such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS), have advanced 3D reconstruction and static scene segmentation, semantic segmentation in 4D dynamic scenes remains relatively underexplored. Key difficulties include maintaining semantic consistency across space and time, as well as the lack of richly annotated multi-view datasets for dynamic environments. This thesis addresses this gap by introducing MUVOD, a multi-view video object segmentation dataset featuring 17 real-world scenarios with 4D-consistent object masks for 459 instances across 73 categories. A benchmark is established for evaluating 4D segmentation methods. Furthermore, we propose a novel 4D segmentation approach based on 3DGS, extending static object segmentation to dynamic scenes by incorporating a learnable Object Identity Encoding into the reconstruction pipeline. The method supports photorealistic RGB and mask rendering at arbitrary viewpoints and timepoints, offering an efficient and portable solution for 4D scene understanding and editing.La vidéo volumétrique permet des représentations 3D dynamiques de scènes réelles, offrant une exploration multi-vues et une navigation naturelle en six degrés de liberté (6 DoF). Pour une interaction véritablement immersive, la segmentation sémantique est essentielle afin de permettre une compréhension des objets au niveau individuel. Bien que les méthodes récentes de synthèse de vues nouvelles, telles que les Neural Radiance Fields (NeRF) et le 3D Gaussian Splatting (3DGS), aient considérablement fait progresser la reconstruction 3D et la segmentation de scènes statiques, la segmentation sémantique dans des scènes dynamiques 4D reste relativement peu explorée. Les principaux défis incluent le maintien de la cohérence sémantique dans l’espace et le temps, ainsi que le manque de jeux de données multi-vues richement annotés pour les environnements dynamiques. Cette thèse répond à ces limitations en introduisant MUVOD, un jeu de données de segmentation d’objets en vidéo multi-vues, couvrant 17 scènes réelles avec des masques 4D cohérents pour 459 instances réparties sur 73 catégories. Un benchmark est proposé pour l’évaluation des méthodes de segmentation 4D. Nous proposons également une nouvelle approche de segmentation 4D fondée sur 3DGS, qui étend la segmentation d’objets statiques aux scènes dynamiques en intégrant un vecteur d’identité d’objet apprenable dans le pipeline de reconstruction. Cette méthode permet le rendu photoréaliste en RGB ainsi que la génération de masques d’objets à n’importe quel point de vue et instant temporel, offrant ainsi une solution efficace et portable pour la compréhension et l’édition de scènes 4D
Éthique interdisciplinaire pratique et appliquée: hybridation pédagogique
No full textInternational audience.This article aims to open the debate on the categorization and distribution of dimensions that can play a role inthe engineering process of hybridization of educational devices. The first part allows us to engage inhermeneutical work and more particularly to question what hybridization is called. A second presentshybridization in the service of teaching and learning with a first part dealing with the criteria, dimensions andtypologies of the devices and a second addressing the training engineering processes. After the presentation ofthe methodology which is based on an abductive approach and the T !O model, we present a distribution of thedimensions of the Hy-Sup project and the properties of the dimensions that result from it. The last part poses,given the distribution of the dimensions identified previously, some theoretical and practical projections onwhat could be an interdisciplinary ethics of hybrid teaching and learning for the 21st century
Adventure in France and Ireland through Literature and Culture
No full textInternational audienceThis collection examines the role that adventure has played and continues to play in French and Irish society and culture in areas as diverse as literature, theatre, war journalism, painting, photography, sport and diplomatic dining. The three types of adventure as defined by the French philosopher Vladimir Jankélévitch are reflected in the various essays. Mortal adventure can be found in the chapters on war journalism and painting, sport and in the pioneer work of the first women photographers and reporters. Aesthetic adventure is depicted in the chapters on the work of Joseph Kessel, Colm Tóibín, Colum McCann, Jan Carson, Anna Burns, Paddy Bushe, John McGahern, Gustave Flaubert and Brian Friel, who have all developed innovative and challenging styles in their attempts to depict aspects of the human condition. Finally, Love adventure is a strong leitmotif in several essays also. We are traversing a time of transformation and transition, a time in which climate change, health crises, wars, and the re-emergence of populism as a force across the planet, have brought us to a crossroads in our history. This collection illustrates how the creative and imaginative power of history, literature and art can help us decipher and redefine our place in this new world and assist us to navigate the challenging adventure(s) that may await us
Droit de préemption urbain et certificat d'urbanisme : le miroir aux alouettes ?
No full textInternational audienceCour administrative d'appel de Marseille, juge des référés, 17 septembre 2025, no 24MA02429, SCI Julien, inédi
(Invited) Microelectronics skills and jobs for the future, in line with digital advancements
No full textInternational audiencea)purposeDigital technology is evolving rapidly in our society. This evolution seems to be very useful for some of our daily activities in a wide range of applications or in the field of research and development, but it also leads to a considerable and permanent increase in the amount of data transferred and processed, more particularly in the context of artificial intelligence tools. Given that all operations in this field are primarily based on electronic equipment and materials, this societal domain is driving exponential growth in the consumption of mainly electrical energy. Within a decade, this could lead to a deleterious global lock-in, with digital activity absorbing the entire global production of electrical energy. Societal behaviors and technologies must change and evolve. To achieve this, new skills and know-how are needed to stimulate innovation in the production of billions of connected objects every year. These correspond to new professions, mainly in microelectronics. Because electronics is at the heart of all equipment, the microelectronics sector has become a priority worldwide. b)design/methodology/approachThe methodology is based on an analysis of the evolution of data transfer worldwide, on the consumption of associated equipment such as data centers. More specifically, it takes account the electrical losses associated with transoceanic optical fiber cables, elementary devices such as nanoelectronic components, as well as circuits and systems, but also the architecture of integrated circuits, systems and software structures. All these products are not at all optimized in the current situation. In addition to these analyses, we need to focus on acquiring the knowledge and know-how needed to meet the various challenges of improvement.c)theoretical base The theoretical basis is derived from the analysis of numerous weaknesses in the technology and design of integrated circuits, as well as the hybrid and complex equipment involved in digital applications, for which initial improvement solutions already exist. These solutions come from research and development laboratories in France, Europe and around the world. Several typical examples will be given, bearing in mind that a new approach needs to be developed. In the case of the software involved in many computer applications, continuous implementation over the past forty years has led to a huge number of bytes, resulting in high power consumption. New, much more concise versions should be developed to reduce the number of gigabytes of each connected object software. The same approaches can be applied to circuit architecture and elementary nanoelectronics devices.d)results or expected results The setting of a national programme in France, France 2030, in agreement and coordinated with a European project, Chips Act, has supported several projects in order to face the challenges. More especially, a project was applied by the “National network for Higher Education in Microelectronics and nanotechnologies, GIP-CNFM. Approved for 5 years and entitled “Engineering for Innovative and Strategic Training in Microelectronics”, INFORISM is devoted to form technicians, engineers and doctors in microelectronics with the target to contribute to the innovation and to apply the approaches that can decrease the energy consumption via many techniques that were suggested by the microelectronics community of researchers that is also involved. The second family of the expected result, should lead to minimize the use of natural resources, knowing that microelectronics industry involves around sixty natural elements, a large amount of them being rare, but also to minimize the influence on the climate.e)originality/valueIts originality lies in the organization of training platforms shared by several academic institutions in the same region and coordinated by a national network. This network enables equipment and operating costs to be pooled, and exchanges of best practice and training to be organized at national level. This practice, acquired by future technicians, engineers and doctors, leads to originality and innovation, thanks in particular to the national educational seminars organized every two years as part of the network. The ultimate aim is to significantly increase the pool of new skills and jobs needed by industry, research and education in this field.f)practical implicationsThe practical implications are to increase knowledge, skills and know-how at all diploma levels, from technicians to engineers and PhDs, in order to meet the needs of French and European industry in the field of microelectronics and nanotechnologies, and to avoid the impasse in electrical energy consumption that is likely to occur as early as the early 30s
(Invited) Challenges of digital world based on ULSI and TFT technologies involving new neuromorphic approaches, Invited paper,
No full textScientific Advisory Committee MemberInternational audienceThe worldwide development of communication and data exchange systems, as well as research into environmental protection, have strongly encouraged the development of digital tools. The current trend is for this growth to accelerate, thanks to the emergence of numerous new applications, crypto-currencies such as blockchains (Bitcoin, Litecoin or Ethereum), and artificial intelligence (AI). The latter approach opens up a wide spectrum of applications in science and technology, but also in writing capabilities such as chatgpt or data processing for commercial or political purposes. The transparency of the operation of these tools and application on the user's side makes us forget that the associated energy consumption is growing exponentially and that we could reach a global dead end in less than 10 years, with the electrical consumption of digital technology exceeding the current world electrical production.These digital tools are physically dependent on electronics and microelectronics, the latter constituting the associated hardware. The solution is to reduce the energy consumption of all electronics, whether analog or digital, power, high-frequency, highly integrated or large-area, but also by modifying the calculation principle in a new algorithm based on neuromorphic similarity. Previous studies [1-2] have highlighted the technological approach of devices that can combine new semiconductor materials and a 3D architecture involving the stacking of ULSI and TFT circuits. These devices are expected to significantly reduce electrical power consumption and improve sustainability in terms of natural resource limitation. The proposed article focuses more on electronic circuits and their architecture, which should curb the considerable increase in power consumption expected as a result of the massive increase in processed data, particularly in artificial intelligence applications.To this end, a good expected solution is to develop a nano-inspired computing task associated with a neuromorphically inspired physical architecture, comprising equivalent neurons, axons or electronic synapses.Early studies were based on the dynamics of neural models that were encoded in software and computed on general-purpose digital hardware. Unfortunately, this technique turned out to require highly complex circuitry, a time-consuming computation process and, consequently, a huge consumption of electrical power. The second technique involves developing an analog hardware approach involving ULSI artificial neurons and synapses. This technique considerably reduces power consumption, but the amount of data processed remains very high.Another approach involves silicon artificial neurons (SAN) based on biophysical model. The most well-known model is the Morris-Leca model that electronically simulates the transport of potassium and sodium elements in the real biological neuron, generating a spike-shaped signal as in a brain. As a result, the energy consumption per spike can be as low as a femtojoule that is promising for future AI systems thanks to an adapted combination of ULSI and TFT technologies for analog circuits and of 3D memories.All these improvements will only be possible if we develop a pool of skills and know-how in the field of microelectronics and nanotechnologies, capable of driving innovation. Challenges that are just as important and complementary to those of a technical nature must therefore be met. As already mentioned, there is a growing skills deficit in the industry, with jobs in high shortage [3]. It is urgent to increase the pool of specialists in the field by innovating on the content of knowledge, and on the pedagogical approach focused on knowledge and know-how. Awareness and attractiveness actions must be carried out at the high school and student levels in order to attract good candidates who will become future actors of the field. One of the challenges of this action is to convince the youth to study in engineering and more especially in microelectronic field. The last part of the paper is dedicated to this challenge and to the actions underway at the French and European levels