Search

This document constitutes the central deliverable of the IO2 ‘Best practices and guidelines for research-based teaching’. It addresses the topic of the research-teaching nexus and provides guidelines for individual lecturers meant to assist them in developing both well-elaborated views on the research-teaching nexus and putting these views into practice through the inclusion of their own ongoing research into ongoing teaching in the form of research-based teaching courses.

International audience; Alcohol use disorders (AUD) is a major public health concern. General practitioners (GP) must play a key role in identifying this disorder and offering appropriate interventions. The authors conducted a survey among French GP in the Provence – Alpes – Côte-d’Azur (PACA) region to better understand their practices regarding AUD. Random sampling was used to enrol 101 GP in a 15-minute survey. A Computer-Assisted Telephone Interview exploring demographic and professional characteristics was conducted using a questionnaire. One third (31.7%) of the participants systematically addressed alcohol use with their patients whereas six (5.9%) never addressed it. Logistic regression analyses showed that after adjustment for the number of AUD patients followed up (odds ratio [OR] = 1.92; 95% confidence interval [CI] [1.19, 3.08]) and GP interest in addiction medicine (5.41 [2.17, 13.40]), GP who systematically screened patients for AUD were more likely to accept controlled drinking as a therapeutic goal (5.41 [2.17, 13.33]), and to perceive patient denial of AUD as a major barrier to care (1.39 [0.60, 3.22]). GP providing care for AUD were more likely to monitor tobacco cessation (9.08 [2.60, 31.64]) and to prescribe opioid maintenance treatment (7.35 [2.52, 21.41]). Alcohol screening is insufficient in general practice in France. Providing updated guidelines is essential to foster experience in this field among GP.

National audience

A neural language model can be conditioned into generating descriptions for images by providing visual information apart from the sentence prefix. This visual information can be included into the language model through different points of entry resulting in different neural architectures. We identify four main architectures which we call init-inject, pre-inject, par-inject, and merge. We analyse these four architectures and conclude that the best performing one is init-inject, which is when the visual information is injected into the initial state of the recurrent neural network. We confirm this using both automatic evaluation measures and human annotation. We then analyse how much influence the images have on each architecture. This is done by measuring how different the output probabilities of a model are when a partial sentence is combined with a completely different image from the one it is meant to be combined with. We find that init-inject tends to quickly become less influenced by the image as more words are generated. A different architecture called merge, which is when the visual information is merged with the recurrent neural network's hidden state vector prior to output, loses visual influence much more slowly, suggesting that it would work better for generating longer sentences. We also observe that the merge architecture can have its recurrent neural network pre-trained in a text-only language model (transfer learning) rather than be initialised randomly as usual. This results in even better performance than the other architectures, provided that the source language model is not too good at language modelling or it will overspecialise and be less effective at image description generation. Our work opens up new avenues of research in neural architectures, explainable AI, and transfer learning., Comment: 145 pages, 41 figures, 15 tables, Doctoral thesis

Natural Language Inference (NLI) is the task of determining the semantic relationship between a premise and a hypothesis. In this paper, we focus on the {\em generation} of hypotheses from premises in a multimodal setting, to generate a sentence (hypothesis) given an image and/or its description (premise) as the input. The main goals of this paper are (a) to investigate whether it is reasonable to frame NLI as a generation task; and (b) to consider the degree to which grounding textual premises in visual information is beneficial to generation. We compare different neural architectures, showing through automatic and human evaluation that entailments can indeed be generated successfully. We also show that multimodal models outperform unimodal models in this task, albeit marginally., Comment: Proceedings of the 12th International Conference on Natural Language Generation (INLG 2019), 11 pages, 5 figures

This paper addresses the sensitivity of neural image caption generators to their visual input. A sensitivity analysis and omission analysis based on image foils is reported, showing that the extent to which image captioning architectures retain and are sensitive to visual information varies depending on the type of word being generated and the position in the caption as a whole. We motivate this work in the context of broader goals in the field to achieve more explainability in AI., 10 pages, 4 figures This publication will appear in the Proceedings of the First Workshop on Shortcomings in Vision and Language (2018). DOI to be inserted later

Image caption generation systems are typically evaluated against reference outputs. We show that it is possible to predict output quality without generating the captions, based on the probability assigned by the neural model to the reference captions. Such pre-gen metrics are strongly correlated to standard evaluation metrics., 13 pages, 6 figures This publication will appear in the Proceedings of the First Workshop on Shortcomings in Vision and Language (2018). DOI to be inserted later

International audience; Introduction Alcohol use disorder (AUD) is a major public health concern worldwide. In France, only 10% of people with AUD (PWAUD) receive medical care. General practitioners (GP) are one of the main entry points for AUD care. The present ongoing study, entitled ASIA (Access to Care and Indifference toward Alcohol, Acces aux Soins et Indifference a l'Alcool in French), aims to improve knowledge about factors associated with access to care for AUD by exploring related GP and PWAUD practices, experiences and perceptions. Methods and analysis The ASIA project is an ongoing cross-sectional multisite study based on a complementary mixed-method approach (quantitative and qualitative) using a convergent parallel design. The double-perspective design of the study will enable us to collect and compare data regarding both PWAUD and GP points of view. For the PWAUD quantitative study, 260 PWAUD will be interviewed using a telephone-based questionnaire. For the qualitative study, 36 PWAUD have already been interviewed. The GP quantitative study will include 100 GP in a 15min survey. Fifteen GP have already participated in semistructured interviews for the qualitative study. Logistic regression will be used to identify predictors for access to care. With respect to data analyses, qualitative interviews will be analysed using semantic analysis while quantitative logistic regression will be used for quantitative interviews. Ethics and dissemination This study was approved by the CNIL (French National Commission on Informatics and Liberties) (approval reference number: C16-10, date of approval: 17 July 2017), the CCTIRS (Advisory Committee on Information Processing in Material Research in the Field of Health) and the CEEI (Evaluation and Ethics Committee) (approval reference number: 16-312, date of approval: 8 July 2016) of INSERM (French National Institute of Health and Medical Research). Results from ASIA will be disseminated in peer-reviewed publications, conference presentations, reports and in a PhD thesis.

We present in this paper our team LCTMALTA’s submission to the RepEval 2017 Shared Task on natural language inference. Our system is a simple system based on a standard BiLSTM architecture, using as input GloVe word embeddings augmented with further linguistic information. We use max pooling on the BiLSTM outputs to obtain embeddings for sentences. On both the matched and the mismatched test sets, our system clearly beats the shared task’s BiLSTM baseline model., peer-reviewed

Le chapitre français de l'ISKO a comme préoccupation majeure, depuis sa création en 1996, la question de l'organisation des connaissances. Cette thématique a été traitée sous l’angle des structures d’organisation, des outils techniques de médiation mais aussi des formes et mécanismes de partage des connaissances. Ces questions sont au cœur des processus de production, de structuration et d’accès à l’information. Dans la continuité de ces réflexions, la 9ème édition du colloque ISKO-France vise à répondre aux problématiques liées à la pertinence de l’information par la contextualisation dans les différentes phases de ces processus. La démocratisation des technologies de l’information et de la communication, en particulier avec Internet, permet à un plus grand nombre de producteurs et diffuseurs de mettre à disposition des informations, comme elle offre à un plus grand nombre d’utilisateurs d’y avoir accès. Or, la production, la mise à disposition et l’accès à l’information sont liés à l’utilisation de ces informations, rendant les problématiques liées à la pertinence de l’information primordiales.