226280 research outputs found
Sort by
L'IA en action : voir comme une machine
No full textNational audiencePreviewThe article explores the rise of image recognition as an emblematic area of applied artificial intelligence. This technology, based on deep learning and massive analysis of visual data, is transforming many sectors: industry (quality control), health (medical diagnosis), security (monitoring), trade (personalization) or human resources (behavioral assessment).We show here that while these uses make it possible to automate formerly human tasks, they are nevertheless based on unincarnated intelligence: algorithms "see" without understanding, detect without feeling, classify without awareness of the context.This observation raises major ethical, organizational and managerial issues, especially in terms of responsibility, algorithmic biases and surveillance at work. Managers are thus called upon to play a pivotal role in the rational integration of these tools, ensuring both their operational efficiency and their social acceptability. The article is part of a thematic series designed to illuminate the different areas of application of AI, highlighting each time the need for enlightened human management in the face of powerful but meaningless systems. ContentArtificial intelligence in action - see as a machine https://management-datascience.org/articles/55033/AperçuL'article explore l'essor de la reconnaissance d'image comme domaine emblématique de l'intelligence artificielle appliquée. Cette technologie, fondée sur l'apprentissage profond et l'analyse massive de données visuelles, transforme de nombreux secteurs : industrie (contrôle qualité), santé (diagnostic médical), sécurité (surveillance), commerce (personnalisation) ou encore ressources humaines (évaluation comportementale).On montre ici, que si ces usages permettent d'automatiser des tâches autrefois humaines, ils reposent néanmoins sur une intelligence non incarnée : les algorithmes "voient" sans comprendre, détectent sans ressentir, classifient sans conscience du contexte.Ce constat soulève des enjeux éthiques, organisationnels et managériaux majeurs, notamment en matière de responsabilité, de biais algorithmiques et de surveillance au travail. Les managers sont ainsi appelés à jouer un rôle pivot dans l'intégration raisonnées de ces outils, en assurant à la fois leur efficacité opérationnelle et leur acceptabilité sociale. L'article s'inscrit dans une série thématique destinée à éclairer les différents domaines d'application de l'IA, en soulignant à chaque fois la nécessité d'un pilotage humain éclairé face à des systèmes puissants mais aveugles au sens. ContenuL'intelligence artificielle en action -voir comme une machine https://management-datascience.org/articles/55033/</div
On K-stability of Fano's last Fanos
No full textWe study K-stability of smooth Fano threefolds of Picard rank 2 and degree 22 which can be obtained by blowing up a smooth complete intersection of two quadrics in P^5 along a conic. We also describe the automorphism groups of these threefolds
Le site antique de Font-de-Leu, sur le domaine de Calissanne (Lançon-de-Provence,13). Nouvelles découvertes en basse vallée de l'Arc
No full textInternational audienceUne occupation antique a été mise au jour au lieu-dit de Font-de-Leu, sur ledomaine de Calissanne (Lançon-de-Provence, 13), au cours d’une opérationde diagnostic. Bien que les données récoltées soient lacunaires en raison ducontexte de découverte, l'étude pluridisciplinaire dont elles ont fait l'objet apermis d'identifier un site à vocation agricole occupé entre le IIe s. av. et leIIIe s. ap. J.-C., dont le développement peut être mis en regard avec celui duterritoire de la basse vallée de l'Arc, dans lequel il se trouve. Des cheminsen évolution sont présents à Font-de-Leu, et le dernier état est couplé à unponceau remarquablement bien conservé, qui comprend dans sa mise enœuvre une stèle funéraire portant une inscription gallo-grecque, inédite parles anthroponymes cités. L'étude du site et du territoire dans lequel il s'intègreest l'occasion d'apporter de nouvelles connaissances et de soulever denouvelles interrogations quant à la dynamique de l'occupation de ce secteu
Auto-édition : une multiplication de livres générés par IA ?
No full textInternational audienceL’auto-édition est présente de longue date dans le monde du livre, même si beaucoup ont vu en elle un phénomène marginal. Voici pourtant que cette pratique solitaire rencontre un succès jamais atteint dans le passé. Les plateformes d’auto-édition avaient déjà facilité et accéléré la possibilité de se publier en toute autonomie, mais par leur simplicité et leur rapidité, les intelligences artificielles (IA) génératives permettent désormais de générer un livre en quelques clics
A Perception Study of Standard Southern British English and Standard Scottish English: Explicit and Implicit Attitudes
No full textInternational audienceIn a previous study (Campana, 2024), we tested how British accents are perceived by British people. The results showed that the Received Pronunciation (RP) and Standard Scottish English (SSE) are the two most appreciated accents, perceived as more pleasant than other British varieties. In the present study, we wish to test the representation of those two varieties via a perception test using both explicit and implicit methods. Indeed, social psychology has shown that humans have two types of reasoning. One which is controlled and slow, explicit, and can be tested through surveys using Likert scales. The other which is automatic, fast and unconscious, implicit, and which will be tested thanks to an implicit association test (Campbell-Kibler, 2012).We propose to duplicate on Standard Southern British English (SSBE) and Standard Scottish English (SSE) a previous study by Mauchand & Pell (2022) in which they tested implicit and explicit attitudes about Parisian French and Quebec French. We will present the test in detail and expose the results. The test will be run on 40 participants (20 SSBE speakers and 20 SSE speakers). The first part of the test will be explicit: the participants will answer questions like “On a 5-point scale, how [pleasant/friendly/educated, etc.] do you think the [English accent / Scottish accent] is?”. The second part of the test will consist in the implicit association test. The participants will first see written words on a computer with positive or negative valences and will have to click on a left button for “pleasant” or a right button for “unpleasant” (block 1). During the whole test, left will thus always be associated to “pleasant” and right to “unpleasant”. Then the participants will be divided into 2 balanced groups (10 SSBE and 10 SSE speakers). The first group will first listen to 3 simple sentences (He’s wearing a coat / She’s riding a bike / They have two horses) produced by 4 SSBE and 4 SSE speakers (n = 24 sentences) and will have to click on the left button if they recognize SSE and on the right button for SSBE (block 2a). The second group will have to click on the left button for SSBE and on the right for SSE (block 2b). Then in a 3rd and 4th block the groups will be presented with both the written words and the audio sentences: the left button for a “pleasant” word or an “SSE” sentence and the right button for an “unpleasant” word or an “SSBE” sentence in the first group (blocks 3a and 4a); and the left button for a “pleasant” word or an “SSBE” sentence and the right button for an “unpleasant” word or an “SSE” sentence in the second group (blocks 3b and 4b), see table 1. The reaction times will be measured. Only the 4th block (a & b) will be analysed, the 3rd block being a trial. We hypothesize that a difference will appear between the explicit and the implicit tests. In Mauchand & Pell (2022)’s study, Canadian participants showed no preference while French participants preferred their own accent. Quebec French was judged more friendly by the French participants. In our study, we expect participants to judge SSE more friendly and SSBE more educated (Moore, 2015) in the explicit test, but we think SSBE might emerge positively in the implicit test.This study will shed light on language attitude (see Mckenzie & Carrie, 2018) and will enable us to better understand how people perceive accents consciously and unconsciously
Exploring the nuances of reduction in conversational speech: lexicalized and non-lexicalized reductions
No full textInternational audienceIn spoken language, a significant proportion of words are produced with missing or underspecified segments, a phenomenon known as reduction. In this study, we distinguish two types of reductions in spontaneous speech: lexicalized reductions, which are well-documented, regularly occurring forms driven primarily by lexical processes, and non-lexicalized reductions, which occur irregularly and lack consistent patterns or representations. The latter are inherently more difficult to detect, and existing methods struggle to capture their full range.We introduce a novel bottom-up approach for detecting potential reductions in French conversational speech, complemented by a top-down method focused on detecting previously known reduced forms. Our bottom-up method targets sequences consisting of at least six phonemes produced within a 230 ms window, identifying temporally condensed segments, indicative of reduction.Our findings reveal significant variability in reduction patterns across the corpus. Lexicalized reductions displayed relatively stable and consistent ratios, whereas non-lexicalized reductions varied substantially and were strongly influenced by speaker characteristics. Notably, gender had a significant effect on non-lexicalized reductions, with male speakers showing higher reduction ratios, while no such effect was observed for lexicalized reductions. The two reduction types were influenced differently by speaking time and articulation rate. A positive correlation between lexicalized and non-lexicalized reduction ratios suggested speaker-specific tendencies.Non-lexicalized reductions showed a higher prevalence of certain phonemes and word categories, whereas lexicalized reductions were more closely linked to morpho-syntactic roles. In a focused investigation of selected lexicalized items, we found that “tu sais” was more frequently reduced when functioning as a discourse marker than when used as a pronoun + verb construction. These results support the interpretation that lexicalized reductions are integrated into the mental lexicon, while non-lexicalized reductions are more context-dependent, further supporting the distinction between the two types of reductions
How does EU non-binding law contribute to the scientific concept of biomedical innovations?
No full textInternational audienceWhereas EU binding law rather uses the expression “biomedical technology” than “biomedical innovation”, the EU legislator seems less hesitant to use the term "innovation" in EU non-binding law. Non-binding EU law plays a significant role in shaping the scientific concept of biomedical innovations and in reflecting on the interactions between innovation and regulation.Firstly, EU non-binding law offers a broader definition of biomedical innovations by emphasizing the central role of society. While reaffirming the dual medical and economic goals of biomedical innovations, it also stresses the importance of public acceptance, aligning closely with the characteristics of biomedical technologies as defined in EU binding law.Secondly, a key finding from the analysis of EU non-binding law is its nuanced approach to the notion of innovation, which sets it apart from binding law in this field. The distinction between incremental and disruptive innovations not only enhances the conceptualization of biomedical innovations but also provides a robust foundation for contemplating the regulatory implications of these innovations. This distinction highlights the mutual influence between regulation and innovation as it is strategically used to justify regulatory decisions. These uncertainties have facilitated the emergence of the principle of innovation as a new policy tool for shaping EU-level regulation of biomedical innovations, leading to its recognition in applicable EU binding law
Is there a legal concept of biomedical innovations in EU binding law?
No full textInternational audienceEuropean Union law plays a great role in regulating “biomedical innovations” such as gene therapy, regenerative medicine or nanomedicine. Yet, the expression has been increasingly used in the literature and has emerged as a concept of science although not being homogenous. Considering it as a concept of legal science, we are defining biomedical innovation as an innovative product, procedure or technique based on biological elements of human or animal origin which entails high or unknown risks, and which serves a medical goal as well as economic competitiveness.Focusing on the biological component of biomedical innovations, and building on the characteristics of legally defined advanced therapy medicinal products, this paper explores if the biomedical innovation’s concept of legal science exists as a legal concept in applicable EU binding law, i.e. if EU law is considering, and/or defining biomedical innovations beyond the legal categories it embeds on the basis of identified and shared characteristics that would deserve to be regulated according to the EU legislator.Although it concludes that the expression “biomedical technologies” would better correspond to the EU legislator’s use, the analysis of EU binding law confirms the relevance of our definition of the biomedical innovation’s concept of legal science
Bouleversement démographique, bouleversement des finances sociales ?
No full textInternational audienc
Bacterial Hsp90 promotes virulence factor production through maintenance of NRPS megaenzymes
No full textAbstract Pathogenic bacteria produce virulence factors critical to host infection. Here, we demonstrate the crucial role of the bacterial Hsp90 chaperone in the production of two major virulence factors, the colibactin genotoxin in Escherichia coli and the pyoverdine siderophore in Pseudomonas aeruginosa . Colibactin, a hybrid polyketide/non-ribosomal peptide (PK-NRP), and pyoverdine, a non-ribosomal peptide (NRP), are metabolites produced by complex biosynthetic pathways involving large cytoplasmic enzymes called megasynthases. Using comparative proteomics, we found that megasynthase abundance was markedly reduced in hsp90 deletion mutants of E. coli and P. aeruginosa compared to wild-type strains. This reduction was independent of transcriptional or translational regulation. We further revealed an interplay between Hsp90 and the HslUV protease in controlling megasynthase levels. Remarkably, we found that Hsp90 stabilizes additional NRP and PK-NRP megasynthases, suggesting a general role for Hsp90 as a chaperone of these enzymes. These findings open new avenues for enhancing the biosynthesis of complex metabolites for biotechnological applications through proteostasis modulation, and may also have implications for combating bacterial infections