A5048549686- Multimodal Machine Learning Applications
- Natural Language Processing Techniques
- Topic Modeling
- Advanced Image and Video Retrieval Techniques
- Healthcare Systems and Practices
- Data-Driven Disease Surveillance
- Domain Adaptation and Few-Shot Learning
- Language, Linguistics, Cultural Analysis
- Cultural Insights and Digital Impacts
- Competitive and Knowledge Intelligence
- linguistics and terminology studies
- Linguistics, Language Diversity, and Identity
- Face recognition and analysis
- Semantic Web and Ontologies
- Diverse multidisciplinary academic research
- Biomedical Text Mining and Ontologies
- Viral Infections and Outbreaks Research
- Substance Abuse Treatment and Outcomes
- Video Analysis and Summarization
- Health, Medicine and Society
- Business Strategy and Innovation
- Influenza Virus Research Studies
- Translation Studies and Practices
- Text Readability and Simplification
- Information Technology and Learning
Santé Publique France
2013-2024
Institut de Recherche pour le Développement
2018-2024
Economic & Social Sciences, Health Systems & Medical Informatics
2014-2024
Inserm
2018-2024
Service de Santé des Armées
2014-2024
Institut Méditerranéen des Sciences de l'Information et de la Communication
2019-2024
University of Malta
2017-2023
Centre de recherche en Epidémiologie et Santé des Populations
2023
Aix-Marseille Université
2018
ARPE PACA
2018
Abstract When a recurrent neural network (RNN) language model is used for caption generation, the image information can be fed to either by directly incorporating it in RNN – conditioning ‘injecting’ features or layer following ‘merging’ features. While both options are attested literature, there as yet no systematic comparison between two. In this paper, we empirically show that not especially detrimental performance whether one architecture another. The merge does have practical...
Image captioning has evolved into a core task for Natural Language Generation and also proved to be an important testbed deep learning approaches handling multimodal representations. Most contemporary rely on combination of convolutional network handle image features, recurrent encode linguistic information. The latter is typically viewed as the primary “generation” component. Beyond this high-level characterisation, CNN+RNN model supports variety architectural designs. dominant in...
The past few years have witnessed renewed interest in NLP tasks at the interface between vision and language. One intensively-studied problem is that of automatically generating text from images. In this paper, we extend to more specific domain face description. Unlike scene descriptions, descriptions are fine-grained rely on attributes extracted image, rather than objects relations. Given no data exists for task, present an ongoing crowdsourcing study collect a corpus images taken `in...
Multilingual language models such as mBERT have seen impressive cross-lingual transfer to a variety of languages, but many languages remain excluded from these models.In this paper, we analyse the effect pre-training with monolingual data for low-resource that is not included in -Maltese -with range set ups.We conduct evaluations newly pre-trained on three morphosyntactic tasks -dependency parsing, part-of-speech tagging, and named-entity recognition -and one semantic classification task...
The widespread under-screening and under-treatment of alcohol use disorder (AUD) contributes to its health socioeconomic burden. We conducted a mixed-methods (qualitative qualitative) study in people with (PWAUD) explore their expectations, as well barriers levers AUD care.Individuals AUDIT > 15 (N = 179) were interviewed using computer-assisted interviews several medical non-medical sites (e.g., bars) (quantitative substudy). also semi-structured face-to-face 36 PWAUD Using logistic...
Recent work has shown evidence that the knowledge acquired by multilingual BERT (mBERT) two components: a language-specific and language-neutral one. This paper analyses relationship between them, in context of fine-tuning on tasks – POS tagging natural language inference which require model to bring bear different degrees knowledge. Visualisations reveal mBERT loses ability cluster representations after fine-tuning, result is supported from identification experiments. However, further...
On 6 July 2018, the Center for Epidemiology and Public Health of French Armed Forces was informed an outbreak acute gastroenteritis among customers a dining facility at military base in Brittany, France. A total 200 patients were reported out population 1700 (attack rate: 12%). The symptoms mainly lower digestive tract disorders occurred rapidly after lunch on 5 (median incubation period: 3.3 h), suggesting toxin-like pathogenic process. case-control survey carried (92 cases 113 controls)....
Occupational exposure to blood and body fluids (BBFs) is a hazard of many occupations, particularly hospital-based health care providers. However, non-hospital-based workers (HCWs) other personnel not providing are also at risk. To present the results surveillance accidental BBFs among French military between 2007 2009, comparing different occupational categories. The study population included all subjected BBF in non-hospital setting. was defined as any percutaneous (needlestick injury,...
The success of cross-lingual transfer learning for POS tagging has been shown to be strongly dependent, among other factors, on the (typological and/or genetic) similarity low-resource language used testing and language(s) in pre-training or fine-tune model. We further unpack this finding two directions by zooming a single language, namely Persian. First, still focusing we run an in-depth analysis behaviour Persian with respect closely related languages that appear benefit from To do so,...
Alcohol use disorders (AUD) is a major public health concern. General practitioners (GP) must play key role in identifying this disorder and offering appropriate interventions. The authors conducted survey among French GP the Provence – Alpes Côte-d'Azur (PACA) region to better understand their practices regarding AUD. Random sampling was used enrol 101 15-minute survey. A Computer-Assisted Telephone Interview exploring demographic professional characteristics using questionnaire. One third...
Propagation Phase Contrast Synchrotron Microtomography (PPC-SRμCT) is the gold standard for non-invasive and non-destructive access to internal structures of archaeological remains. In this analysis, virtual specimen needs be segmented separate different parts or materials, a process that normally requires considerable human effort. Automated SEgmentation Imaging (ASEMI) project, we developed tool automatically segment these volumetric images, using manually samples tune train machine...
Résumé The main function of the Defense Health Service is to preserve health projected forces. In this context, a watch duty was defined and entrusted observation Unit French Forces Institute Tropical Medicine. To complete mission, three documentation databases were created broadcasted. Their updating, depending on innovation detection, requires process documentary awareness in six phases : definition topics be supervised, selection sources, gathering, analysis, reformatting diffusion...
Résumé Un service de veille scientifique est amené à collecter une masse importante documents qualité variable. Pour identifier les plus pertinents d’entre eux, l’Unité sanitaire l’Institut médecine tropicale du santé des armées (IMTSSA) a élaboré grille d’analyse en sept étapes. En tenant compte la rigueur contenu et, dans le cas d’un article, facteur d’impact revue publication et niveau preuve l’étude, cette permet d’estimer valeur document. Marc Tanti , collaboration avec Christian Hupin...
When designing a neural caption generator, convolutional network can be used to extract image features. Is it possible also use language model sentence prefix features? We answer this question by trying different ways transfer the recurrent and embedding layer from an generator. find that generators with transferred parameters perform better than those trained scratch, even when simply pre-training them on text of same captions dataset will later on. best models (in terms perplexity) do not...
In recent years, the number of data circulating on Internet has exploded causing phenomenon "Big Data". health field, mass also become bloated. From a study conducted within french medical service, this article aims to explore construction knowledge generated by "Sanitary Big It analyzes creation value made and limitations " The result is an exploitation polymorphic mainly from global programs for infectious disease surveillance including PromedMail Sentiweb. shows use clinical trials...