HAL Université de Tours
Not a member yet
55851 research outputs found
Sort by
« Un océan de détails : la peinture d’histoire de John Martin et sa réception par les contemporains »
No full textInternational audienc
« Les suites d’Homère existent-elles ? Un essai d’introduction au colloque de Tours »
No full textInternational audienc
A longitudinal person-centered investigation of individual employees’ team boosting behaviors
No full textInternational audienceThis person-centered investigation sought to identify the nature of employees' profiles of boosting behaviors (i.e., energizing, mood-enhancing, and uniting behaviors seeking to support the functioning of their team). We also examined the stability of these profiles over time, and of their associations with a series of predictors (i.e., availability expectations, social challenge stressors, telepressure, and harassment) and outcomes (i.e., affective commitment to the organization and to coworkers, and organizational citizenship behaviors directed towards individuals). We identified five distinct profiles among a sample of 415 employees who completed the same set of measures twice across a time interval of three months: Very Low Boosting Behaviors, High Boosting Behaviors, Low Boosting Behaviors, Very High Boosting Behaviors, and Average Boosting Behaviors. These profiles were moderately to highly stable over time. Social challenge stressors and telepressure were associated with a lower likelihood of membership into the Very Low Boosting Behaviors profile. All outcomes also differed across profiles, with the most adaptive outcomes being associated with the Very High Boosting Behaviors profile and the most maladaptive outcomes being associated with the Very Low Boosting Behaviors profile
Reproduction de l’expérience : alignement des mots par affinage des embeddings sur des corpus parallèles (NLP)
No full textDans cette étude,nous avons cherché à reproduire l’expérience de l’article "Word Alignment by Fine-tuning Embeddings on Parallel Corpora" de Zi-YiDou et Graham Neubig, qui propose une méthode innovante pour l’alignement des mots dans des corpus parallèles. En exploitant les embed-dings contextuels pré-entrainés issus de modèlesde langage multilingues, l’article présente une ap-proche qui affine ces embeddings sur des don-nées parallèles afin d’améliorer la précision desalignements (Luong et al. 2015; Garg et al. 2019;Zenkel et al. 2020). Nous avons adapté cetteapproche en concentrant notre travail sur un cor-pus parallèle anglais-français, axé sur des thé-matiques médicales, et en utilisant différents out-ils et bibliothèques d’apprentissage automatique.Ce processus nous permet d’explorer les perfor-mances de plusieurs techniques d’alignement demots, en mettant particulièrement l’accent sur lesméthodes d’alignement basées sur les modèlesd’apprentissage profond et les embeddings con-textuels. Nos résultats, en comparaison avec lesméthodesclassiquescommeFastAlign(Dyeretal.,2013) et Eflomal (Ostling and Tiedemann, 2016),visent à démontrer l’efficacité des modèles pré-entrainés dans la gestion d’alignements (Sabet etal. 2020) multilingues robustes, même dans desconfigurations zéro-shot
Strict inequality between the time constants of first-passage percolation and directed first-passage percolation
No full textIn the models of first-passage percolation and directed first-passage percolation on Z^d, we consider a family of i.i.d random variables indexed by the set of edges of the graph, called passage times. For every vertex x ∈ Z^d with nonnegative coordinates, we denote by t(0, x) the shortest passage time to go from 0 to x and by \vec{t}(0, x) the shortest passage time to go from 0 to x following a directed path. Under some assumptions, it is known that for every x ∈ R^d with nonnegative coordinates, t(0, ⌊nx⌋)/n converges to a constant µ(x) and that \vec{t}(0, ⌊nx⌋)/n converges to a constant \vec{µ}(x). With these definitions, we immediately get that µ(x) ≤ \vec{µ}(x). In this paper, we get the strict inequality µ(x) < \vec{µ}(x) as a consequence of a new exponential bound for the comparison of t(0,x) and \vec{t}(0,x) when ||x|| goes to \infty. This exponential bound is itself based on a lower bound on the number of edges of geodesics in first-passage percolation (where geodesics are paths with minimal passage time)
Les peintres et la médaille au Salon dans la seconde moitié du XIXe siècle
No full textInternational audienc
Différences de processus anticipatoires entre l'initiation de la marche et du pas
No full textInternational audienc
A teleclinic for Neuro-psychological Assessment in the Metaverse
No full textInternational audienceMetaverse technologies allow new tele-medicine services and applications. Our interdisciplinary team conducted a user-centered design process, and implemented an immersive teleclinic for clinicians to carry out neuro-psychological testing sessions with distant patients. A session is a mix of discussions and desk-based or independent test activities, accompanied and monitored by the clinician with various dashboards. We carried out a preliminary evaluation focusing on usability, presence and acceptability for healthy participants playing the role of patients and practitioners. Results shows all the participants could project into using the system, but practitioners were more reluctant, due so some usability issues, and insufficient quality of current technology to observe patients
SDHI : ces pesticides agricoles qui inquiètent les scientifiques
No full textNational audienceSDHI : ces pesticides agricoles qui inquiètent les scientifique
Evaluation of serum mid-infrared spectroscopy as new prognostic marker for first-line bevacizumab-based chemotherapy in metastatic colorectal cancer
No full textInternational audienceBackground and aims: Bevacizumab-based chemotherapy is a recommended first-line treatment for metastatic colorectal cancer (mCRC). Robust biomarkers with clinical practice applicability have not been identified for patients with this treatment. We aimed to evaluate the prognostic yield of serum midinfrared spectroscopy (MIRS) on patients receiving first-line bevacizumab-based chemotherapy for mCRC. Methods: We conducted an ancillary analysis from a multicentre prospective study (NCT00489697). All baseline serums were screened by attenuated total reflection method. Principal component analysis and unsupervised k-mean partitioning methods were performed blinded to all patients' data. Endpoints were progression-free survival (PFS) and overall survival (OS). Results: From the 108 included patients, MIRS discriminated two prognostic groups. First group patients had significantly lower body mass index ( p = 0.026) and albumin levels ( p < 0.001), and higher levels of angiogenic markers, lactate dehydrogenase and carcinoembryonic antigen ( p < 0.001). In univariate analysis, their OS and PFS were shorter with respective medians: 17.6 vs 27.9 months ( p = 0.02) and 8.7 vs 11.3 months ( p = 0.03). In multivariate analysis, PFS was significantly shorter (HR = 1.74, p = 0.025) with a similar trend for OS (HR = 1.69, p = 0.061). Conclusion: By metabolomic fingerprinting, MIRS proves to be a promising prognostic tool for patients receiving first-line bevacizumab-based chemotherapy for mCRC.</div