HAL Portal Univ-Côtedazur
Not a member yet
95391 research outputs found
Sort by
Human–Bird Interactions Across Time and Space in a Bronze Age City: The Case of Tell Atchana, Alalakh (Amuq Valley, Turkey)
No full textInternational audienceABSTRACT Birds have played both subsistence and symbolic roles in past human societies, with their significance evolving alongside sedentary lifestyles and agriculture. Although Neolithic settlements in Western Asia primarily relied on domesticated mammals, birds remained a marginal resource, their importance varying by region. This study investigates bird remains from Bronze Age Alalakh (Tell Atchana) to understand their role in an urbanizing society across shifting political and environmental conditions in the Amuq Valley. Avian remains recovered from Tell Atchana (2007–2012) were analyzed using zooarchaeological methods, including taxonomic identification via comparative osteology and further classification through traditional morphometrics. Quantitative analyses assessed skeletal representation, diversity indices, and abundance patterns across excavated time periods and areas. Contextual association with food preparation equipment, cut marks, and skeletal representation provided further insight into human–bird interactions. Results reveal a strong prevalence of waterfowl, comprising over 75% of identified birds, with mallard ( Anas platyrhynchos ) and teal ( Anas crecca ) as the most abundant bird species. Diachronic variation is evident across major bird taxa, with peak diversity and abundance observed in the Late Bronze Age I, along with a substantial and increasing frequency of migratory birds. These patterns may correspond with broader hydrological and cultural shifts in the region, as well as greater reliance on wild animal resources at times of political unrest. The emphasis on wing elements, particularly in pigeons and mallards, suggests these birds held specific ritual or symbolic significance. Further study is needed to clarify the multifaceted meanings birds held for Alalakh inhabitants, particularly regarding distinctions between consumption, management, and ritual use. Continued investigation into human–bird interactions at the site will contribute to broader discussions on environmental adaptation and cultural practices in early urban societies
Simulating Shock Solutions for Hyperbolic Conservation Laws with Physics-Informed Neural Networks
No full textPhysics-informed neural networks represent a new paradigm for modeling complex systems, relying on both data and physical laws (e.g., partial differential equations) for the training. This duality is particularly interesting when little data is available. However, this formulation relies on smooth representations to describe the solutions, so it is not well suited to solving problems where discontinuities occur, as can be the case with nonlinear hyperbolic conservation laws. In this work, we explore how to solve this type of problems using multiple networks, describing on the one hand the behavior of the system in regular regions and on the other hand the jump conditions between these regions. The methods developed are applied to the Burgers equation with shocks
Distilling Foundation Models for Robust and Efficient Models in Digital Pathology
No full textInternational audienceIn recent years, the advent of foundation models (FM) for digital pathology has relied heavily on scaling the pre-training datasets and the model size, yielding large and powerful models. While it resulted in improving the performance on diverse downstream tasks, it also introduced increased computational cost and inference time. In this work, we explore the distillation of a large foundation model into a smaller one, reducing the number of parameters by several orders of magnitude. Leveraging distillation techniques, our distilled model, H0-mini, achieves comparable performance to large FMs at a significantly reduced inference cost on HEST and EVA public benchmarks. Additionally, we conduct robustness analyses on the PLISM-WSI dataset and a multi-scanner, multi-staining private breast cancer cohort. We demonstrate that our distilled model reaches excellent robustness to variations in staining and scanning conditions, significantly outperforming other state-of-the-art models. This opens new perspectives to design lightweight and robust models for digital pathology, without compromising on performance. We publicly release H0-mini along with plismbench, the first robustness benchmark of pathology foundation models based on the PLISM dataset
Q²Forge: Minting Competency Questions and SPARQL Queries for Question-Answering Over Knowledge Graphs
No full textInternational audienceThe SPARQL query language is the standard method to access knowledge graphs (KGs). However, formulating SPARQL queries is a significant challenge for non-expert users, and remains time-consuming for the experienced ones. Best practices recommend to document KGs with competency questions and example queries to contextualise the knowledge they contain and illustrate their potential applications. In practice, however, this is either not the case or the examples are provided in limited numbers. Large Language Models (LLMs) are being used in conversational agents and are proving to be an attractive solution with a wide range of applications, from simple question-answering about common knowledge to generating code in a targeted programming language. However, training and testing these models to produce high quality SPARQL queries from natural language questions requires substantial datasets of question-query pairs. In this paper, we present Q²Forge that addresses the challenge of generating new competency questions for a KG and corresponding SPARQL queries. It iteratively validates those queries with human feedback and LLM as a judge. Q²Forge is open source, generic, extensible and modular, meaning that the different modules of the application (CQ generation, query generation and query refinement) can be used separately, as an integrated pipeline, or replaced by alternative services. The result is a complete pipeline from competency question formulation to query evaluation, supporting the creation of reference query sets for any target KG
The necessary caution of public administrations in adopting new tools derived from artificial intelligence
No full textInternational audienceThe arrival of artificial intelligence (technological evolution) is bringing new perspectives to french public administrations. The French case is interesting to study for several reasons. Firstly, it is one of the OECD countries where the role of the public sphere is one of the most important in terms of government spending as a percentage of GDP. Secondly, France is also a country where the use of artificial intelligence in public services is modest. AI is only present in internal public service processes or at the level of service design and delivery, and not in improving the formulation of public policies, unlike in countries such as Canada, UK or United States.This reluctance on the part of government agencies can be explained by a lack of understanding of the new technology and by their natural legal caution (the French precautionary principle) in the face of novelty. Thus, ethic is often developed to explain why administrations are so reluctant to commit to AI. Other principles are evoked at the highest level to circumscribe this caution. These include the principles of human primacy, performance, equity and non-discrimination, transparency, safety, environmental sustainability and strategic autonomy. Generative AI is being experimented with in a number of ministries, including the Ministry of Economy and Finance. While the results of this superficial AI, based as it is on a specific business dimension, are very encouraging, their impact remains limited. For example, the “Llamandement” project at the Public Finances General Directorate was able to automate 3 out of 4 stages in the process of handling parliamentary amendments during the examination of the finance act. This has improved performance, without compromising service quality. This results in improved performance, without any deterioration in service quality. However, some experiments have not had the same positive effects (“Service public +”, “je donne mon avis”). This study highlights the advantages and limitations of the choices made in France and in a few countries (UK, US and Canada) considered to be more advanced in the implementation of AI within central government
Hesitations and gradual adoption of artificial intelligence by French public administrations
No full textInternational audienceAll over the world, public administrations are embracing artificial intelligence. The arguments in its favor are well known, including greater quantitative efficiency, the use of automation, and stricter enforcement of rules. So why have French central government administrations opted for a measured approach to AI development? Numerous reports (from ministries, the Council of State, agencies, etc.) highlight this unusual approach, which is very much in line with the precautionary principle that prevails in France when analyzing innovations.This chapter attempts to explain why French administrations are taking longer to integrate AI into their activities. There are many obstacles: cultural and social resistance, institutional, technical and political barriers, and ethical and societal issues. Based on experiments with generative AI and machine learning conducted by the French Ministry of Economy and Finance and on experiences abroad, it examines the gradual adoption of AI in central government and its consequences for the quality of public services. This study highlights the advantages and limitations of the choices made in France and in a few countries considered to be more advanced in the implementation of AI within central government
Well-quasi-orders on embedded planar graphs
No full textThe central theorem of topological graph theory states that the graph minor relation is a well-quasi-order on graphs. It has far-reaching consequences, in particular in the study of graph structures and the design of (parameterized) algorithms. In this article, we study two embedded versions of classical minor relations from structural graph theory and prove that they are also well-quasi-orders on general or restricted classes of embedded planar graphs. These embedded minor relations appear naturally for intrinsically embedded objects, such as knot diagrams and surfaces in . Handling the extra topological constraints of the embeddings requires careful analysis and extensions of classical methods for the more constrained embedded minor relations. We prove that the embedded version of immersion induces a well-quasi-order on bounded carving-width plane graphs by exhibiting particularly well-structured tree-decompositions and leveraging a classical argument on well-quasi-orders on forests. We deduce that the embedded graph minor relation defines a well-quasi-order on plane graphs via their directed medial graphs, when their branch-width is bounded. We conclude that the embedded graph minor relation is a well-quasi-order on all plane graphs, using classical grids theorems in the unbounded branch-width case
Intermittency assessed through a model of kurtosis–skewness relation in MHD in fast dynamo regimes
No full textInternational audienceIntermittency as it occurs in fast dynamos in the magnetohydrodynamics (MHD) framework is evaluated through the examination of relations between normalized moments at third order (skewness ) and fourth order (kurtosis ) for both the velocity and magnetic field, and for their local dissipations. As investigated by several authors in various physical contexts such as fusion plasmas (Krommes 2008 Phys. Plasmas 15 , 030703), climate evolution (Sura & Sardeshmukh 2008 J. Phys. Oceano. 38 , 639-647), fluid turbulence or rotating stratified flows (Pouquet et al. 2023 Atmosphere 14 , 01375), approximate parabolic laws emerge whose origin may be related to the applicability of intermittency models to their dynamics. The results analyzed herein are obtained through direct numerical simulations of MHD flows for both Taylor–Green and Arnold–Beltrami–Childress forcing at moderate Reynolds numbers, and for up to turn-over times. We observe for the dissipation , an evaluation that varies with the field, the forcing and when filtering for high-skewness intermittent structures. When using the She & Lévêque (1994) Phys. Rev. Lett. 72 , 336-339 intermittency model, one can compute analytically; we then find , clearly differing from a (strict) parabolic scaling, a result consistent with the numerical data
Long-term observation of fluid venting features in the Amazon Fan
No full textInternational audienceDiscoveries of gas venting from the deep seafloor attract growing attention from the scientific community and the energy industry, given their implications for the energy transition and greenhouse gas emissions. Understanding the dynamics of gas hydrate systems and associated exudation processes is essential for assessing their potential environmental and economic impacts. The Amazon River culminates in one of the world’s largest deep-sea fans, offering a natural laboratory to study gas migration and expulsion within a rapidly-deposited and gravitationally collapsing depocentre. Gas venting has been documented within an upper slope compressional belt during a decade-long observational study involving campaigns in 2013 and 2023, which acquired hydroacoustic data and core samples that included gas hydrates. This study integrates these datasets with exploration 2D and 3D seismic data to investigate seafloor gas venting features and their connections to active fault systems. Over the 10-year observation period, within the same area of 1549 km² (water depths 900-1800m), water column gas flares increased in number, with 34 new flares identified in 2023; 17 flares observed in 2013 disappeared, while 13 remained active in 2023. The flares rise from seafloor mounds, and in some cases depressions, interpreted as mud volcanoes and possibly pockmarks. These seafloor vents are commonly associated with acoustically chaotic subsurface vertical zones interpreted as fluid escape conduits. In the case of mud volcanoes, conduits of kilometric vertical extent rise from anticlines and are associated with deformation of surrounding layers and extrusion of material onto the seafloor. Most venting structures lie above and pass through bottom simulating reflection (BSR) patches that cross-cut the tops of buried or seafloor anticlines; the BSR in places exhibits ‘pluming’ behavior, rising toward seafloor vents. The seafloor with the upper slope compressional belt is offset by both normal faults, observed above the crests of buried anticlines, and by thrust-faults within the anticlines which extend downward to shale detachments in upper Miocene and older formations. Bright spot reflections, often observed adjacent to faults, highlight zones of gas migration along these structures. Our findings underscore the widespread distribution of upper slope fluid vents linked to complex subsurface geological structures including active folds and faults. The temporal variability of gas venting, characterized by the emergence, persistence, and disappearance of gas flares, highlights the dynamic nature of these processes and their significance for understanding methane cycling and its implications