HAL Portal AU (University of Avignon)
Not a member yet
35713 research outputs found
Sort by
There are significant differences among artificial intelligence large language models when answering scientific questions
No full textInternational audienceIntroduction: This study investigates the efficacy of large language models (LLMs) for generating accurate scientific responses through a comparative evaluation of five prominent free models: Claude 3.5 Sonnet, Gemini, ChatGPT 4o, Mistral Large 2, and Llama 3.1 70B. Methods: Sixteen expert scientific reviewers assessed these models in terms of depth, accuracy, relevance, and clarity. Results: Claude 3.5 Sonnet emerged as the highest scoring model, followed by Gemini, with notable variability among the other models. Additionally, retrieval-augmented generation (RAG) techniques were applied to improve LLM performance, and prompts were refined to improve answers. The results indicate that although LLMs such as Claude 3.5 Sonnet have potential for scientific tasks, other models may require more development or additional prompt engineering to reach comparable accuracy. Reviewers' perceptions of artificial intelligence (AI) utility and trustworthiness showed a positive shift after evaluation. However, ethical concerns, particularly with respect to transparency and disclosure, remained consistent. Discussion: The study highlights the need for structured frameworks for evaluating LLMs and ethical considerations essential for responsible AI integration in scientific research. These findings should be interpreted with caution, as the limited sample size and domain-specific focus of the exam questions restrict the generalizability of the results
Educational Nudges and Teacher Agency in France’s School of Trust Paradigm
No full textInternational audienceThis article presents preliminary findings from an educational collaboration research study conducted in a middle school within the framework of the School of Trust paradigm. In classroom settings, teachers employ educational nudges as mediating tools to align their pedagogical activities with the principles of educational benevolence, fostering pupil self-regulation and volition. Grounded in activity theory, this case study utilized data from situated observations and reflective video-confrontation interviews with teachers to explore the impact of educational nudges on classroom dynamics. While self-reported data on student volition and self-regulation is limited, this study demonstrates how these tools enhance teacher agency and contribute to the transformation of traditional schooling practices. The findings suggest that educational nudges not only support student agency, but also empower teachers to innovate within the School of Trust paradigm, paving the way for a more benevolent and effective educational environment
Étude génétique et linguistique de l’incipit de la Recherche : une première approche
No full textInternational audienceThis document, which presents the ANR Cré@lame at the study day "Representing Writing. Tools, Techniques, Applications October 17, 2025 – University of Quebec (Canada)," outlines a study of the opening lines of Marcel Proust's Recherche using an interdisciplinary approach (literature, lexicometry, cognitive psychology, criticism, genetics) to the processes of writing and creation and their use in monitoring and training in creative writing. It also presents the ethical and legal issues raised by the recording, analysis, and use of processes by key loggers. Translated with DeepL.com (free version)Ce document, support de la présentation de l'ANR Cré@lame lors de la journée d'étude "Représenter l’écriture. Outils, techniques, exploitations 17 octobre 2025 – Université du Québec (Canada)", esquisse une étude de l'incipit de la Recherche de Marcel Proust dans une approche interdisciplinaire (littérature, psychologie cognitive, critique, génétique) des processus d'écriture et de création et de leur utilisation dans le suivi et la formation à l'écriture de création. Il présente également les questions éthiques et juridiques que posent l'enregistrement, l'analyse et l'utilisation des processus par des key loger
Comment structurer un message d'alerte pour maximiser sa compréhension par la population: Retours d’expériences scientifiques et perspectives
No full textInternational audienc
Remote sensing and spatial modelling. Applications to the surveillance and control of mosquito-borne diseases
No full textSource Agritrop Cirad (https://agritrop.cirad.fr/615033/)International audienceMosquitoes are vectors of many disease-causing pathogens, including malaria, dengue, chikungunya, and yellow fever. According to the World Health Organization, these vector-borne diseases account for several hundred thousand deaths annually. They also cause zoonoses, such as Rift Valley fever and West Nile fever. In this context, the development of operational tools to support surveillance and control strategies is essential—not only in countries of the Global South, where mosquito-borne diseases are most prevalent in tropical and subtropical regions, but also in the countries of the North, where the establishment of invasive species such as the tiger mosquito is increasing the risk of disease emergence. To address these challenges, Earth observation imagery offers valuable potential: the spatial distribution and seasonal dynamics of mosquito populations are closely linked to climatic factors (such as temperatures, rainfall, and humidity) and environmental variables (such as the presence of water bodies and vegetation), many of which can be monitored through satellite data. Numerous recent studies have led to the development of innovative methods that combine remote sensing with spatial modelling to predict the spatial and temporal dynamics of vector mosquitoes and associated diseases. Moving beyond proof-of-concept, some of these approaches have given rise to operational tools and processing chains that are now actively used by public health authorities and vector control agencies. This book, intended for students, researchers, and public health professionals, offers a synthesis of current research and operational tools in the field
Deploying Disaster-Resilient Service Function Chains Using Adaptive Multi-Path Routing: Network Function Virtualization (NFV) is a new technology that deploys network services and functions as software components in data centers and cloud environments. One of its key applications is Service Function Chain (SFC), which chains a set of Virtual Network Functions (VNFs) in a specific order to deliver a desired service. However, deploying NFV and SFC networks faces challenges, particularly in terms of disaster resiliency. This encompasses natural disasters and hardware failures, which can disrupt network operations and lead to service interruption or degradation across an entire disaster zone (DZ). Therefore, designing NFV and SFC networks that can withstand disasters while providing high levels of service availability and reliability is important. This paper presents a new method for protecting SFCs using adaptive multi-path routing. The proposed Multi-path Protection (MP) method has the advantage of reducing the amount of reserved bandwidth on backup paths by distributing SFC traffic over multiple DZ-disjoint working paths. The problem being addressed involves VNFs placement, routing SFCs, and implementing protection mechanisms. The objective is to minimize network resource consumption, including both the bandwidth used by request routing paths and the computing resources for VNF execution. To solve this multi-dimensional optimization problem, a path-adaptive and flow-based integer linear program (ILP) is proposed to provide the optimal solution in sall-size network settings. We also propose a heuristic approach that offers the near-optimal solution in a time-efficient way. Comprehensive simulation results show that the proposed MP strategy outperforms traditional Dedicated Protection (DP) in terms of bandwidth and processing resource consumption, resulting in a significant gain up to 20%.
No full textInternational audienc
How food processing and storage affect Bacillus cereus presence in pea protein-based products?
No full textInternational audienc
Does the temporal variation of leaf terpene and moisture content trigger leaf flammability over time?
No full textInternational audienceBackground – aims It is widely assumed that plant flammability in the Mediterranean region peaks during the summer fire season. We currently lack data that could evaluate these assumptions and have not assessed the mechanisms, e.g. fuel moisture content (FMC) or terpenes, that might drive these patterns. Methods To determine the mechanistic drivers of species flammability, we used leaf burning experiments coupled with foliar chemical analyses focusing on Aleppo pine (Pinus halepensis) and three introduced cypresses commonly found at the wildland–urban interface (WUI) in southeastern France. Key results Terpenes, FMC and flammability varied over time and across the species studied, with contrasting patterns for each. Rare correlations between FMC and flammability occurred, in only one season and differing among species, while correlations between flammability and terpene compounds were diverse. The best flammability drivers were terpenes (mainly diterpenes), often changing among and within seasons, and their effect on flammability also differed. Overall, FMC was not a significant explanatory parameter of leaf flammability. Conclusions – implications Highlighting the temporal variation between flammability and its drivers revealed that species flammability could also be enhanced by terpenes outside the fire season; this should be accounted for in fire prevention, especially at the WUI
Exposure to PFAS: One Health approach in the anthropised area of Fos-Berre. Study of contamination and toxicity of PFAS in drinking water and analytical developments for lichen biomonitoring of atmospheric contamination
No full textInternational audienc
Screening Sourdough Starter Cultures from Yeast and Lactic Acid Bacteria Isolated from Mexican Cocoa Mucilage and Coffee Pulp for Bread Quality Improvement
No full textInternational audienceThis study aimed to identify and evaluate yeasts and lactic acid bacteria (LAB) isolated from Mexican cocoa mucilage (Theobroma cacao) and coffee pulp (Coffea arabica) for their potential use as sourdough starter co-cultures to improve bread quality. Functional screens included assessments of amylolytic, proteolytic, and phytase activities, CO2 production, acidification capacity, and exopolysaccharide (EPS) synthesis. Saccharomyces cerevisiae YCTA13 exhibited the highest fermentative performance, surpassing commercial baker’s yeast by 52.24%. Leuconostoc mesenteroides LABCTA3 showed a high acidification capacity and EPS production, while Lactiplantibacillus plantarum 20B3HB had the highest phytase activity. Six yeast–LAB combinations were formulated as mixed starter co-cultures and evaluated in sourdough breadmaking. The B3Y14 co-culture (LABCTA3 + YCTA14) significantly improved the bread volume and height by 35.61% and 17.18%, respectively, compared to the commercial sourdough starter, and reduced crumb firmness by 59.66%. Image analysis of the bread crumb revealed that B3Y14 enhanced the crumb structure, resulting in greater alveolar uniformity and a balanced gas cell geometry. Specifically, B3Y14 showed low alveolar regularity (1.16 ± 0.03) and circularity (0.40 ± 0.01), indicating a fine and homogeneous crumb structure. These findings highlight the synergistic potential of selected allochthonous yeast and LAB strains in optimizing sourdough performance, positively impacting bread texture, structure, and quality