Search

Limited data resources are the current problem for neural machine translation of low resources languages. In this paper we propose to use multilingual embedding as a method of representing words as input to a neural machine translator. This approach is then compared to word representation approaches in monolingual context used in the literature for neural machine translation. The results with multilingual embedding on a dataset of 7187 pairs of French-Ewondo parallel verses of the bible are promising because they are better than those obtained with the representation approaches used so far. We obtained 8.77% of blue for the unsupervised multilingual embedding,5.34% for the semi-supervised version and finally 4.92% for the word embedding with the skipgram architecture.; Les ressources de données limitées sont le problème actuel de la traduction automatique neuronale des langues à faibles ressources. Dans cet article, nous proposons d'utiliser le plongement multilingue comme méthode de représentation des mots en entrée d'un traducteur neuronal. Cette approche est ensuite comparée aux approches de représentation de mots en contexte monolingue utilisées dans la littérature pour la traduction automatique neuronale. Les résultats avec le plongement multilingue sur un jeu de données de 7187 paires de versets parallèles français-éwondo de la bible sont prometteurs car ils sont meilleurs que ceux obtenus avec les approches de représentation utilisées jusqu'à présent. Nous avons obtenu 8,77% de bleu pour le plongement multilingue non supervisé, 5,34% pour la version semi-supervisée et enfin 4,92% pour le plongement de mots avec l'architecture skipgram.

International audience; The traditional algorithms that extract the gradual patterns often face the problem of managing the quantity of mined patterns, and in many applications, the calculation of all these patterns can prove to be intractable for the user-defined frequency threshold. Moreover, the concept of gradualness is defined just as an increase or a decrease variation. Indeed, a gradualness is considered as soon as the values of the attribute on both objects are different. This does not take into account the level of variation. Then, the variation of 10^-6 is considered as the same way as that of 10^6. As a result, numerous quantities of patterns extracted by traditional algorithms can be presented to the user, although their gradualness (due to the small variation) could be only a noise in the data. To address this issue, this paper suggests introducing the gradualness threshold from which to consider an increase or a decrease variation. In contrast to the literature approaches, the proposed approach takes into account the user's preferences on the gradualness threshold. The user knowledge could be used to fix the value of gradualness threshold. The proposed algorithm makes it possible to extract gradual patterns on certain databases where state-of-the-art gradual patterns mining algorithms fail due to too large search space. Moreover, results from an experimental evaluation on real databases show that the proposed algorithm is scalable, efficient, and can eliminate numerous patterns that do not verify specific gradualness requirements to show a small set of patterns to the user.

Introduction Detecting safety signals attributed to a drug in scientific literature is a fundamental issue in pharmacovigilance. The constant increase in the volume of publications requires the automation of this tedious task, in order to find and extract relevant articles from the pack. This task is critical, as serious Adverse Drug Reactions (ADRs) still account for a large number of hospital admissions each year. Objectives The aim of this study is to develop an augmented intelligence methodology for automatically identifying relevant publications mentioning an established link between a Drug and a Serious Adverse Event, according to the European Medicines Agency (EMA) definition of seriousness. Methods The proposed pipeline, called LiSA (for Literature Search Application), is based on three independent deep learning models supporting a precise detection of safety signals in the biomedical literature. By combining a Bidirectional Encoder Representations from Transformers (BERT) algorithms and a modular architecture, the pipeline achieves a precision of 0.81 and a recall of 0.89 at sentences level in articles extracted from PubMed (either abstract or full-text). We also measured that by using LiSA, a medical reviewer increases by a factor of 2.5 the number of relevant documents it can collect and evaluate compared to a simple keyword search. In the interest of re-usability, emphasis was placed on building a modular pipeline allowing the insertion of other NLP modules to enrich the results provided by the system, and extend it to other use cases. In addition, a lightweight visualization tool was developed to analyze and monitor safety signal results. Conclusions Overall, the generic pipeline and the visualization tool proposed in this article allows for efficient and accurate monitoring of serious adverse drug reactions from the literature and can easily be adapted to similar pharmacovigilance use cases. To facilitate reproducibility and benefit other research studies, we also shared a first benchmark dataset for Serious Adverse Drug Events detection.

International audience; L’agriculture constitue aujourd’hui l’un des principaux utilisateurs de laressource en eau. Dans ce contexte, de nombreux travaux ont mis en avant l’intérêt descompteurs communicants pour mieux gérer cette ressource. Néanmoins, ces compteursposent des questions en termes d’acceptabilité, ce qui nuit à la diffusion de ces technologiesauprès des agriculteurs. Nous proposons donc de recourir à la simulation à based’agents pour mieux comprendre les dynamiques et les freins à leur diffusion. Contrairementà la grande majorité des travaux sur la diffusion d’innovations qui se limitent à unereprésentation abstraite et simplifiée de ce processus, nous proposons dans cet article unmodèle générique reposant sur la théorie du comportement planifié et sur l’argumentationformelle permettant d’expliquer les raisons du changement d’opinion d’un agent, élémentfondamental pour comprendre la dynamique de diffusion de l’innovation. Chaque agenta ainsi la possibilité d’échanger des arguments avec un autre et de construire son opinionsur une innovation à partir de l’ensemble des arguments qu’il connaît. Les premièresexpérimentations menées à partir de données recueillies auprès des agriculteurs de larivière du Louts montrent une tendance à une plus grande adoption de ces compteurset soulignent l’importance que peuvent avoir les fake news ainsi que certains argumentscritiques sur le processus d’adoption.

International audience; The predominantly agricultural Mekong Delta region will be profoundly affected by the effects of climate change, such as rising temperatures, sea level rise, and salinization. But recent studies show that other factors, such as sand mining and groundwater exploitation, will also largely contribute to future pressures, particularly through land subsidence. Alongside the fight against climate change, this makes it necessary to design strategies and policies for subsidence mitigation, not only at the individual level (mainly farmers), but also at the institutional level (province and region). Indeed, in the Vietnamese Mekong Delta (VMD), local provinces play a crucial role in the choice and implementation of strategies for the exploitation of natural resources, such as water or sand. However, the VMD is also a complex system in which the decisions of one province can have indirect effects on the other 12 provinces, especially because some provinces share common agro-ecological regions and aquifer areas. This chapter aims to explore, through simulation of an integrated model, the economic impact on farmers of the adaptation strategies chosen by the provinces under a number of climatic and economic scenarios. It thus builds on the adaptation options evaluated in Part 3 of the report. In particular, one objective is to understand which conditions of inter-province coordination on groundwater exploitation in common agro-ecological areas could improve the economic situation of farmers. To this end, the LUCAS model (an agent-based model for land-use change adaptation strategies in the context of climate change) [ Truong et al., 2021 ] was extended to take groundwater use by farmers into account, as well as the subsidence dynamics and macroeconomic trends from the GEMMES model. In the updated model, called LUCAS-GEMMES, subsidence-related adaptation strategies chosen at the provincial level and their interactions with individual decisions were also introduced. The simulation is performed with 4 experiments including (i) No provincial adaptation strategies and ignorance of subsidence dynamics; (ii) No adaptation strategies while subsidence impacts the benefit of land-use production; (iii) Individual adaptation strategies combined with the impact of subsidence; (iv) Province and individual adaptation combined with the impact of subsidence.The results show that decisions in early response to low-level subsidence bring many positive results in resource management, such as significantly reducing water use in the dry season, and reducing the vulnerability area from subsidence and climate change. The results also show a decrease in incomes in case of adaptation. Therefore, it is necessary to have investment solutions (technical support in shrimp farming, studying for new upland crops in the dry season) for new land-use types that could compensate for the economic benefits lost when applying policies that restrict people in the use of resources.; La région du delta du Mékong, essentiellement agricole, sera profondément touchée par les effets du changement climatique, tels que la hausse des températures, l’élévation du niveau de la mer et la salinisation. Mais des études récentes montrent que d’autres facteurs, comme l’extraction de sable et l’exploitation des eaux souterraines, contribueront également fortement aux pressions futures, notamment par la subsidence. Au-delà de la lutte contre le changement climatique, il est donc nécessaire de concevoir des stratégies et des politiques d’atténuation de la subsidence, non seulement au niveau individuel (principalement les agriculteurs), mais aussi au niveau institutionnel (province et région). En effet, dans le delta du Mékong vietnamien (DMV), les provinces locales jouent un rôle crucial dans le choix et la mise en œuvre des stratégies d’exploitation des ressources naturelles, comme l’eau ou le sable. Cependant, le DMV est aussi un système complexe dans lequel les décisions d’une province peuvent avoir des effets indirects sur les 12 autres provinces, notamment parce que certaines provinces partagent des régions agro-écologiques et des zones aquifères communes. Ce chapitre vise à explorer, par la simulation d’un modèle intégré, l’impact économique sur les agriculteurs des stratégies d’adaptation choisies par les provinces dans un certain nombre de scénarios climatiques et économiques. Il s’appuie donc sur les options d’adaptation évaluées dans la partie 3 du rapport. En particulier, un objectif est de comprendre dans quelles conditions de coordination entre les provinces sur l’exploitation des eaux souterraines, dans des zones agro-écologiques communes, la situation économique des agriculteurs peut s’améliorer. À cette fin, le modèle LUCAS (un modèle “basé agents” pour les stratégies d’adaptation au changement d’utilisation des terres dans le contexte du changement climatique) [ Truong et al., 2021 ] a été étendu pour prendre en compte l’utilisation des eaux souterraines par les agriculteurs, ainsi que la dynamique de subsidence et les tendances macroéconomiques du modèle GEMMES. Dans le modèle mis à jour, appelé LUCAS-GEMMES, les stratégies d’adaptation liées à la subsidence choisies au niveau provincial et leurs interactions avec les décisions individuelles ont également été introduites. La simulation est réalisée à l’aide de quatre expériences : (i) aucune stratégie d’adaptation provinciale et ignorance de la dynamique de la subsidence ; (ii) aucune stratégie d’adaptation alors que la subsidence a un impact sur le bénéfice de la production de l’utilisation des terres ; (iii) stratégies d’adaptation individuelles combinées à l’impact de la subsidence ; (iv) adaptation provinciale et individuelle combinée à l’impact de la subsidence. Les résultats montrent que les décisions prises en réponse précoce à un affaissement de faible niveau apportent de nombreux résultats positifs dans la gestion des ressources, tels que la réduction significative de l’utilisation de l’eau pendant la saison sèche, puis la réduction de la zone de vulnérabilité à la subsidence et au changement climatique. Les résultats montrent également une diminution des revenus en cas d’adaptation. Par conséquent, il est nécessaire d’avoir des solutions d’investissement (soutien technique dans l’élevage de crevettes, étude de nouvelles cultures de montagne en saison sèche) pour de nouveaux types d’utilisation des terres qui compenseraient les avantages économiques perdus lors de l’application de politiques qui limitent les personnes dans l’utilisation des ressources.

International audience; Nowadays, job recommender systems are more useful in the fight against unemployment due to their strong presence in e-recruitment platforms that are becoming very popular. Most of the recommender systems based on machine learning that recommend jobs to profiles use a vector representation of job offers based on keywords. However, these keywords are results of vectorization which is applied on a collection of documents where each one is a job offer. In this case, each keyword discriminates one job offer from another, whereas it can be preferable that each keyword discriminates one class from another. The aim of this paper is to improve the recommendation of job offers to users, by proposing to apply vectorization on a class-oriented collection of documents in order to obtain more useful keywords for the representation of job offers. In the case of job recommendation, each class-oriented document corresponds to a user profile. Experiments are done on two datasets (Nigam and Minajobs), using TF-IDF and Doc2Vec as vectorization techniques, Naive Bayes and Decision Trees as supervised machine learning models for top-N recommendation, and Precision, MRR and MAP as evaluation metrics. Our results show that, whatever the case, the best performance is always reach by a job recommender system resulting from our contribution. Compared to classic job recommender systems, the improvement rates can go up to 13% and 24% for systems based on Naive Bayes and go up to 55% and 46% for those based on Decision tree, respectively in the Nigam and Minajobs datasets.; Aujourd'hui, les systèmes de recommandation d'emploi sont plus utiles dans la lutte contre le chômage en raison de leur forte présence dans les plateformes de recrutement en ligne qui deviennent très populaires. La plupart des systèmes de recommandation basés sur l'apprentissage automatique qui recommandent des emplois à des profils utilisent une représentation vectorielle des offres d'emploi basée sur des mots-clés. Cependant, ces mots-clés sont les résultats d'une vectorisation appliquée à une collection de documents dont chacun est une offre d'emploi. Dans ce cas, chaque mot-clé discrimine une offre d'emploi par rapport à une autre, alors qu'il peut être préférable que chaque mot-clé discrimine une classe par rapport à une autre. L'objectif de cet article est d'améliorer la recommandation d'offres d'emploi aux utilisateurs, en proposant d'appliquer la vectorisation sur une collection de documents orientée classe afin d'obtenir des mots-clés plus utiles pour la représentation des offres d'emploi. Dans le cas de la recommandation d'emploi, chaque document orienté classe correspond à un profil d'utilisateur. Les expériences sont réalisées sur deux jeux de données (Nigam et Minajobs), en utilisant TF-IDF et Doc2Vec comme techniques de vectorisation, Naive Bayes et Decision Trees comme modèles d'apprentissage automatique supervisés pour la recommandation top-N, et Precision, MRR et MAP comme métriques d'évaluation. Nos résultats montrent que, dans tous les cas, la meilleure performance est toujours atteinte par un système de recommandation d'emploi issu de notre contribution. Par rapport aux systèmes classiques de recommandation d'emploi, les taux d'amélioration peuvent aller jusqu'à 13% et 24% pour les systèmes basés sur Naive Bayes et jusqu'à 55% et 46% pour ceux basés sur Decision tree, respectivement dans les jeux de données Nigam et Minajobs.

International audience; Repetitive tasks are most often tedious; in order to facilitate their execution, program synthesis approaches have been developed. They consist in automatically inferring programs that satisfy the intention of a user. The best known approach in program synthesis is FlashFill integrated into the Excel spreadsheet which allows the processing of character strings. In FlashFill user intent is represented by examples i.e, pairs (input, output). FlashFill explores a very large space of programs and therefore can require a lot of execution time and infer a lot of programs some of which work on given examples but do not capture the user's intent. In this article, we propose two approaches QuickMixte and QuickFill based on blocks which aim to guide the exploration of the program space of FlashFill by enriching the specifications provided by the user. These approaches ask the user to provide associations between subparts of the output and the input to refine the specifications. Experiments carried out on a series of 12 datasets show that QuickMixte and QuickFill make it possible to considerably reduce the program space of FlashFill. We show that with these approaches, it is often possible to give fewer examples than with the original FlashFill algorithm for a larger proportion of correct programs.Keywords : Synthesis of programs, programming by example, manipulation of character strings, repetitive tasks, block approach.; Les tâches répétitives sont le plus souvent fastidieuses ; afin de faciliter leur exécution, les approches de synthèse de programmes ont été développées. Elles consistent à inférer automatiquement des programmes qui satisfont l'intention d'un utilisateur. L'approche la plus connue en synthèse de programmes est FlashFill intégrée au tableur Excel qui permet le traitement des chaînes de caractères. Dans FlashFill l'intention de l'utilisateur est représentée par des exemples i.e, des couples (entrée, sortie). FlashFill explore un très grand espace de programmes et peut donc nécessiter un temps d'exécution important et inférer beaucoup de programmes dont certains fonctionnent sur des exemples donnés mais ne capturent pas l'intention de l'utilisateur. Dans cet article, nous proposons deux approches QuickMixte et QuickFill basées sur les blocs qui visent à guider l'exploration de l'espace de programmes de FlashFill en enrichissant les spécifications fournies par l'utilisateur. Ces approches demandent à l'utilisateur de fournir des associations entre les sous-parties de la sortie et de l'entrée pour affiner les spécifications. Les expérimentations menées sur une série de 12 jeux de données montrent que QuickMixte et QuickFill permettent de réduire considérablement l'espace de programmes de FlashFill. Nous montrons qu'avec ces approches, il est souvent possible de donner moins d'exemples qu'avec l'algorithme FlashFill original pour une plus grande proportion de programmes corrects.

International audience; Top-N recommendation systems are important because they influence user choices on a daily basis on streaming and e-commerce platforms. Recommendation graphs are one of the commonly used approaches for computing top-N recommendations, as they are intuitive and interpretable. The idea is to build a graph from explicit data of users' actions on products, and then apply a random walk algorithm like PageRank to it to propose recommendations. However, implicit data that are not directly perceptible in the history of users' actions, such as latent data resulting from matrix factorization, are not exploited in recommendation graphs. Indeed, at the end of a matrix factorization, users and products are represented in a new space defined by the latent factors, in which a user and an item are close if the item is of the user's taste. In this paper, we propose to exploit the latent information resulting from matrix factorization to enrich recommendation graphs. Experiments are performed on six datasets, using three metrics for evaluating top-N recommendations: Precision, MAP and Hit-ratio. Our results show that the graphs enriched with latent information are better in 97% of the cases compared to the initial graph. The performance improvement can go up to at least 33%, depending on the dataset.; Les systèmes de recommandation top-N sont importants car ils influencent quotidiennement les choix des utilisateurs des plateformes notamment celles de streaming et de commerce électronique. Les graphes de recommandation sont l'une des approches couramment utilisées pour le calcul des recommandations top-N, car ces derniers sont intuitifs et interprétables. Le principe est de construire un graphe à partir des données explicites des actions des utilisateurs sur les produits, puis d'y appliquer un algorithme de marche aléatoire comme le PageRank pour proposer des recommandations. Cependant, les données implicites qui ne sont pas directement perceptibles dans l'historique des actions des utilisateurs à l'exemple des données latentes résultantes de la factorisation matricielle ne sont pas exploitées dans les graphes de recommandation. En effet, à l'issue d'une factorisation matricielle les utilisateurs et les produits sont représentés dans un nouvel espace défini par les facteurs latents, dans lequel un utilisateur et un item sont proches si l'item est du goût de l'utilisateur. Dans cet article, nous proposons d'exploiter les informations latentes résultantes de la factorisation matricielle pour enrichir les graphes de recommandation. Des expérimentations sont effectuées sur six jeux de données, en utilisant trois métriques d'évaluation des recommandations top-N : Précision, MAP et Hit-ratio. Nos résultats montrent que les graphes enrichis par les informations latentes sont meilleurs dans 97% des cas comparé au graphe initial. L'amélioration des performances peut aller jusqu'à 33% au moins, selon le jeu de données.

International audience; Jacovi et al. in 2018 proposed a method for the explanation of Convolutional Neural Networks designed for text classification (Text-CNN). The problem with this method is that it only applies to Text-CNN with no hidden layer in the fully connected part. We will therefore find ourselves limited in terms of accuracy if we want to use this method. In addition, the LRP (Layer-wise Relevance Propagation) method makes it possible to compute the contributions of inputs of neural networks with multiple hidden layers, but cannot be applied directly to Text-CNNs because unlike images where each pixel has a meaning taken alone, each component of a vector representation (word embedding) of a word taken alone has no particular meaning. Therefore, we propose in this article to extend the method of Jacovi et al. by using the LRP method in order to be able to explain Text-CNN whose fully connected layer is deep. The qualitative evaluation suggests that the explanations provided by the proposed method are consistent on text classification problems such as Question-Answering (QA) and Sentiment Analysis. Moreover, the quantitative evaluation shows that the explanations provided are 100% faithful to the model.; Jacovi et al. en 2018 ont proposé une méthode pour l'explication des Réseaux de Neurones Convolutifs conçus pour la classification des textes (Text-CNN). Le problème avec cette méthode est qu'elle ne s'applique que sur les Text-CNN qui n'ont pas de couche cachée dans la partie densément connectée. On se retrouvera donc limité en terme de performance si on veut utiliser cette méthode d'explication. Par ailleurs, la méthode LRP (Layer-wise Relevance Propagation) permet de calculer les contributions des caractéristiques en entrée des réseaux de neurones avec un nombre quelconque de couches cachées, mais ne peut s'appliquer directement sur les Text-CNN car contrairement aux images où chaque pixel a un sens pris seul, chaque composante d'une représentation vectorielle (word embedding) d'un mot prise seule n'a pas de signification particulière. C'est ainsi que nous proposons dans cet article, d'étendre la méthode de Jacovi et al. en mettant à contribution la méthode LRP afin de pouvoir expliquer les Text-CNN dont la partie densément connectée est profonde (possède plusieurs couches). L'évaluation qualitative laisse percevoir que les explications fournies par la méthode proposée sont cohérentes sur les problèmes de classification des textes comme le Question-Answering (QA) ainsi que l'analyse des sentiments. Par ailleurs, l'évaluation quantitative montre que les explications fournies sont fidèles à 100% au modèle.

International audience; Comparison is a powerful tool for decision support. Many measures have been proposed for comparing two metric spaces, but they do not consider the data dispersion in each metric space. Furthermore, no dedicated measure has yet been proposed for comparing two metric spaces containing elements belonging to various subtypes of a global datatype. In order to attenuate the aforementioned limitations, this paper proposes a new technique for comparing two metric spaces. More formally, given a metric space X, histograms are used for performing a gradual analysis of the data dispersion inside the neighborhood of each element of X. This is a refinement of the neighborhoods' analysis realized in an existing work. Then, another existing technique is used for associating one hidden Markov model X with X such that X learns the bin values and the visual shapes of the histograms derived from the instances in X. Meta-data derived from X are then saved as the components of a descriptor vector �⃗ X associated with X. Finally, the comparison between two metric spaces is performed through the comparison of their respective associated descriptor vectors using existing distance or similarity measures between two vectors. The proposed approach inherits the accuracy and the efficiency of the existing techniques on which it relies. Therefore, the experiments realized in this paper are only intended to show how it can be used for comparing particular metric spaces containing geolocations or stars in the celestial sphere.

Trees are among the most studied data structures and several techniques have consequently been developed for comparing two trees belonging to the same category. Until the end of year 2020, there was a serious lack of suitable metrics for comparing two weighted trees or two trees from different categories. The problem of comparing two tree sets was not also specifically addressed. These limitations have been overcome in a paper published in 2021 where a customizable metric based on hidden Markov models have been proposed for comparing two tree sets, each containing a mixture of trees belonging to various categories. Unfortunately, that metric does not allow the use of non metric-dependent classifiers which take descriptor vectors as inputs. This paper addresses this drawback by deriving a descriptor vector for each tree set from the behavior of its associated models after a sufficiently long time. The comparison between two tree sets is then realized by comparing their associated descriptor vectors. Classification experiments carried out on the databases FirstLast-L (FL), FirstLast-LW (FLW) and Stanford Sentiment Treebank (SSTB) respectively showed best accuracies of 99.75%, 99.75% and 87.22%. These performances are respectively 40.75% and 20.52% better than the tree Edit distance respectively for FLW and SSTB. Additional clustering experiments exhibited 54.25%, 98.75% and 75.53% of correctly clustered instances respectively for FL, FLW and SSTB. No clustering was performed in existing work.; Les arbres font partie des structures de données les plus étudiées et plusieurs techniques ont par conséquent été développées pour comparer deux arbres appartenant à la même catégorie. Jusqu'à la fin de l'année 2020, il y avait un sérieux manque de métriques appropriées pour comparer deux arbres pondérés ou deux arbres de catégories différentes. Le problème de la comparaison de deux ensembles d'arbres n'a pas non plus été spécifiquement abordé. Ces limites ont été surmontées dans un article publié en 2021 dans lequel une métrique paramétrable basée sur des modèles de Markov cachés a été proposée pour comparer deux ensembles d'arbres, chacun contenant un mélange d'arbres appartenant à différentes catégories. Malheureusement, cette métrique ne permet pas l'utilisation de classificateurs ne dépendant pas d'une mesure de distance qui prennent généralement des vecteurs en entrées. Le présent article s'attaque à cette limite en proposant un descripteur vectoriel pour chaque ensemble d'arbres à partir du comportement sur une longue période des modèles qui lui sont associés. La comparaison entre deux ensembles d'arbres est alors réalisée en comparant leurs descripteurs vectoriels respectifs. Des expériences en classification menées sur les bases de données FirstLast-L (FL), FirstLast-LW (FLW) et Stanford Sentiment Treebank (SSTB) ont respectivement permis d'obtenir les meilleures performances de 99,75 %, 99,75 % et 87,22 %. Ces performances sont respectivement 40,75% et 20,52% meilleures que la 'Tree Edit distance' respectivement pour FLW et SSTB. Des expériences supplémentaires en clustering ont permis d'avoir respectivement 54,25 %, 98,75 % et 75,53 % instances correctement groupées pour FL, FLW et SSTB. Aucun clustering n'avait été effectué dans les travaux existants.

International audience; At a time when the impacts of climate change and increasing urbanization are making risk management more complex, there is an urgent need for tools to better support risk managers. One approach increasingly used in crisis management is preventive mass evacuation. However, to implement and evaluate the effectiveness of such strategy can be complex, especially in large urban areas. Modeling approaches, and in particular agent-based models, are used to support implementation and to explore a large range of evacuation strategies, which is impossible through drills. One major limitation with simulation of traffic based on individual mobility models is their capacity to reproduce a context of mixed traffic. In this paper, we propose an agent-based model with the capacity to overcome this limitation. We simulated and compared different spatio-temporal evacuation strategies in the flood-prone landlocked area of the Phuc Xa district in Hanoi. We demonstrate that the interaction between distribution of transport modalities and evacuation strategies greatly impact evacuation outcomes. More precisely, we identified staged strategies based on the proximity to exit points that make it possible to reduce time spent on road and overall evacuation time. In addition, we simulated improved evacuation outcomes through selected modification of the road network.

Sparsely endowed languages puts down a performance problem for neural machine translation because these have limited data resources. In this paper we propose to use crosslingual word embedding as a method of words representation in the entrance of neural machine translator. This approach is then compared to the approaches of words representation in monolingual context used in the literature for neural machine translation. The results we get with crosslingual word embedding on a dataset of 5325 words in Ewondo and 6490 in French are promising because they are better than those obtained with the representation approaches used so far. moreover we improve the Blue and Accuracy metrics by 0.357% and 1.369% respectively when for crosslingual words embedding we consider the agglutinant character of Ewondo language by performing manually groupings of words before the tokenization step; Les langues peu dotées posent un problème de performance pour la traduction automatique neuronale car celles-ci disposent de ressources en données limitées. Dans ce papier nous proposons d'utiliser le plongement multilingue comme méthode de représentation de mots en entrée d'un traducteur automatique neuronal. Cette approche est ensuite comparée aux approches de représentation de mots en contexte monolingue utilisées dans la littérature pour la traduction automatique neuronale. Les résultats que nous obtenons avec le plongement multilingue sur un jeu de données de 5325 mots en Ewondo et 6490 en Français sont prometteurs car ils sont meilleurs que ceux obtenus avec les approches de représentations jusqu'ici utilisées. De plus nous améliorons les métriques Bleu et Exactitude de 0.357% et 1.369% respectivement lorsque pour le plongement multilingue nous considérons le caractère aglutinant de l'Ewondo en effectuant manuellement des regroupements de mots avant l'étape de tokenisation. Mots-Clés Plongement multilingue de mots, Réseaux de neurones, Traduction automatique neuronale, Langues faiblement dotées.

International audience; Agent-based simulation has long been used to study the dynamics of adoption and diffusion of innovations. However, the vast majority of these works are limited to an abstract and simplified representation of this process, which does not allow to explain the reasons for the change of opinion of an agent. In order to go further in the explanation of these changes, we present a generic model based on the theory of planned behavior and on formal argumentation. Each agent has the possibility to exchange arguments with another and to build its opinion on an innovation from the set of arguments it knows. An application of the model is proposed to study the adoption of communicating water meters by farmers on the Louts river (South-West of France).

International audience; Artificial Intelligence has opened new doors for customer relationship personalization by capturing life events to tailor front and back-office interactions. Individual bank account data are particularly rich in information on these life events, but few banks have gone beyond its basic use. In this paper, we describe an innovative and original methodological framework to give meaning to bank transactions and make them actionable under operational and regulatory constraints. The approach includes unsupervised methods that limit upstream feature engineering and are based on a global modeling of a customer’s journey through sequence objects.

Technological advances have made high-resolution sequencing of genetic material possible at ever lower cost. In this context, the human microbiome (considered as our second "genome") has demonstrated its great capacity to stratify various human diseases. As a "super-integrator" of patient status, the gut microbiota is set to play a key role in precision medicine. Omics biomarkers identification has become a major goal of metagenomics processing, as it allows us to understand the microbial diversities that induce the patient stratification. There remain many challenges associated with mainstream metagenomics pipelines that are both time consuming and not stand-alone. This prevents metagenomics from being used as "point-of-care" solutions, especially in resource-limited or remote locations. Indeed, state-of-the-art approaches to learning from metagenomics data still relies on tedious and computationally heavy projections of the sequence data against large genomic reference catalogs. In this thesis, we address this issue by training deep neural networks directly from raw sequencing data building an embedding of metagenomes called Metagenome2Vec. We also explore subgroup discovery algorithms that we adapt to build a classifier with a reject option which then delegates samples, not belonging to any subgroup, to a supervised algorithm. Several datasets are used in the experiments to discriminate patients based on different diseases (colorectal cancer, cirrhosis, diabetes, obesity) from the NCBI public repository. Our evaluations show that our two methods reach high performance comparable to the state-of-the-art, while being respectively stand-alone and interpretable.; Les progrès technologiques ont rendu possible le séquençage haute résolution du matériel génétique à faible coût. Dans ce contexte, le microbiome humain a démontré sa grande capacité à stratifier diverses maladies humaines. En tant que "super-intégrateur" de l'état du patient, le microbiote intestinal est appelé à jouer un rôle clé dans la médecine de précision. L'identification de biomarqueurs omiques est devenue un objectif majeur en métagénomique, et permet de comprendre les diversités microbiennes qui induisent la stratification des patients. Il reste de nombreux défis associés aux pipelines de métagénomique courants, qui prennent du temps et ne sont pas autonomes. Cela empêche l'utilisation de la métagénomique comme solution "point-of-care", comme dans les régions à ressources limitées. En effet, l'état de l'art de l'apprentissage à partir de données métagénomiques repose sur des projections lourdes en calcul des séquences sur de grands catalogues de référence génomique. Dans cette thèse, nous abordons ce problème en entraînant des réseaux neuronaux directement à partir des séquences brutes, en construisant un "embeddings" des métagénomes appelé Metagenome2Vec. Nous explorons également des algorithmes de découverte de sous-groupes que nous adaptons en classifieur avec une option de rejet qui délègue ensuite les échantillons n'appartenant à aucun sous-groupe à un algorithme supervisé. Plusieurs jeux de données sont utilisés dans les expériences sur diverses maladies (cancer colorectal, cirrhose, diabète, obésité) à partir du répertoire NCBI. Nos évaluations montrent que nos méthodes atteignent des performances élevées comparables à l'état de l'art.

International audience

International audience; The rice-shrimp farming system is considered as a sustainable and beneficial model for the environment. However, the area of rice-shrimp was increasingly narrowed due to the trend of converting from rice to aquaculture by economic reasons. This paper aims to propose a medium scale land use change model for understanding the land use decision of farmers in adaptation to the environment and climate change. The model integrates a land-use decision making process based on multi-criteria selection where the main factors are land suitability, land convertibility, land use situation of neighbors, and profitability of land use patterns. Concerning the land use data, we used historical land use map in 2005, 2015 and 2019. Shrimp cultivation regions was completed by Landsat satellite image processing. The model has been calibrated by rice-shrimp map in 2015 and has been verified with the rice – shrimp map in 2019 of the My Xuyen district, Soc Trang province, Vietnam. The simulated results show that the rice-shrimp area was increasingly narrowed and has been converted to aquaculture land. In addition, the model tends to show that in a scenario of sea level rise of 15 cm in 2030, the share of rice-shrimp and shrimp tends to rise sharply, which is an important lesson for developing complex adaptive strategies of farmers.

International audience; An approach that is particularly well adapted to study the dynamics of adoption and diffusion of innovations is agent-based simulation. It allows modelers to take into account the complex interactions between actors as well as their heterogeneity. Numerous works have already shown the interest of this method for the study of innovation diffusion processes. However, the vast majority of these works have been limited to an abstract and simplified representation of this process. This very abstract representation does not allow users to understand and explain the reasons for the change of opinion of an agent, which is nevertheless fundamental to understanding the dynamics of innovation diffusion. In order to overcome this limitation, we propose an agent-based model of adoption and diffusion of innovations that uses a structured argumentation framework. An application of this model is proposed to study the diffusion of communicating water meters by farmers on the Louts river (South-West of France) and shows that the introduction of new arguments could impact the adoption process.

PurposeThe overall functioning of this system is based on two approaches: construction and supervision. The first is conducted entirely by the machine, and the second requires the intervention of the designer to collaborate with the machine. The morphological translation of urban rules is sometimes contradictory and may require additional external relevance to urban rules. Designer arbitration assists the artificial intelligence (AI) in accomplishing this task and solving the problem.Design/methodology/approachThis paper provides a method of computational design in generating the optimal authorized bounding volume which uses the best target values of morphological urban rules. It examines an intelligent system, adopting the multi-agent approach, which aims to control and increase urban densification by optimizing morphological urban rules. The process of the system is interactive and iterative. It allows collaboration and exchange between the machine and the designer. This paper is adopting and developing a new approach to resolve the distributed constraint optimization problem in generating the authorized bounding volume. The resolution is not limited to an automatic volume generation from urban rules, but also involves the production of multiple optimal-solutions conditioned both by urban constraints and relevance chosen by the designer. The overall functioning of this system is based on two approaches: construction and supervision. The first is conducted entirely by the machine and the second requires the intervention of the designer to collaborate with the machine. The morphological translation of urban rules is sometimes contradictory and may require additional external relevance to urban rules. Designer arbitration assists the AI in accomplishing this task and solving the problem. The human-computer collaboration is achieved at the appropriate time and relies on the degree of constraint satisfaction. This paper shows and analyses interactions with the machine during the building generation process. It presents different cases of application and discusses the relationship between relevance and constraints satisfaction. This topic can inform a chosen urban densification strategy by assisting a typology of the optimal authorized bounding volume.FindingsThe human-computer collaboration is achieved at the appropriate time and relies on the degree of constraint satisfaction with fitness function.Originality/valueThe resolution of the distributed constraint optimization problem is not limited to an automatic generation of urban rules, but involves also the production of multiple optimal ABV conditioned both by urban constraints as well as relevance, chosen by the designer.

The Internet contains a wealth of information in the form of unstructured texts such as customer comments on products, events and more. By extracting and analyzing the opinions expressed in customer comments in detail, it is possible to obtain valuable opportunities and information for customers and companies. The model proposed by Jebbara and Cimiano. for the extraction of aspects, winner of the SemEval2016 competition, suffers from the absence of lexico-grammatic input characteristics and poor performance in the detection of compound aspects. We propose the model based on a recurrent neural network for the task of extracting aspects of an entity for sentiment analysis. The proposed model is an improvement of the Jebbara and Cimiano model. The modification consists in adding a CRF to take into account the dependencies between labels and we have extended the characteristics space by adding grammatical level characteristics and lexical level characteristics. Experiments on the two SemEval2016 data sets tested our approach and showed an improvement in the F-score measurement of about 3.5%. L'analyse des opinions consiste à extraire des connaissances à partir des commentaires laissés par les utilisateurs à propos d'un produit, service, texte,... L'analyse des opinions basée sur les aspects consiste alors à décomposer le commentaire afin d'extraire les aspects que cet utilisateur a évalué. Le modèle proposé par Jebbara et Cimiano, vainqueur de la compétition SemEval2016 n'extrait pas correctement des aspects composés et ne prend pas en compte les caractéristiques lexico-grammaticales des textes en entrée, ce qui limite aussi ses performances dans la détection des aspects. Nous proposons une amélioration du modèle de Jebbara et Cimiano. en y introduisant des unités CRF afin de prendre en compte les dépendances entre les étiquettes et ajoutant aux entrées du modèle des caractéristiques lexico-grammaticales. Les expérimentations faites sur les deux jeux de données de SemEval2016 ont permis de tester cette approche et montrer une amélioration de la mesure F-score d'environ 3.5%. ABSTRACT. The Internet contains a wealth of information in the form of unstructured texts such as customer comments on products, events and more. By extracting and analyzing the opinions expressed in customer comments in detail, it is possible to obtain valuable opportunities and information for customers and companies. The model proposed by Jebbara and Cimiano. for the extraction of aspects, winner of the SemEval2016 competition, suffers from the absence of lexico-grammatic input characteristics and poor performance in the detection of compound aspects. We propose the model based on a recurrent neural network for the task of extracting aspects of an entity for sentiment analysis. The proposed model is an improvement of the Jebbara and Cimiano model. The modification consists in adding a CRF to take into account the dependencies between labels and we have extended the characteristics space by adding grammatical level characteristics and lexical level characteristics. Experiments on the two SemEval2016 data sets tested our approach and showed an improvement in the F-score measurement of about 3.5%.

International audience; Since the adoption of the Sustainable Development Goals (SDGs), the international community’s efforts to achieve the SDGs have been unevenly distributed. An exploration of the data related to these goals raises the urgent need to mo- nitor them in order to better focus efforts in the least develo- ped countries. We propose here a method based on Machine Learning to overcome the lack of data and the need for dynamic monitoring. This method is based on three prin- ciples : participatory research, context localized process and dynamic model development. An illustrative example is presented in the context of the development of new data sets and prediction models in Senegal, showing the interest of the proposed approach.

https://hal-emse.ccsd.cnrs.fr/emse-03278442/file/actes_CNIA_CH_PFIA2021.pdf; National audience; This contribution introduces the use of convolutional neural networks for the characterization of magnetic anomalies. The developed approach allows one the localization of magnetic dipoles, including counting the number of dipoles, their geographical position, and the prediction of their parameters (magnetic moment, depth, and declination). Subsequently, it will be tested on real data, for example, in the framework of pyrotechnic detection for unexploded ordnance prospection. Finally, an application to other geophysical methods will be considered.; Cette contribution présente l'utilisation des réseaux de neurones convolutifs pour la détection d'anomalies magnétiques. L'approche développée permet la localisation de dipôles magnétiques, avec le comptage du nombre de dipôles, leur position géographique et la prédiction de leurs paramètres (moment magnétique, profondeur et déclinaison). Elle sera ensuite testée sur des données réelles, dans le cadre par exemple, d'une détection pyrotechnique pour la prospection de munitions non explosées, avant d'envisager une application vers d'autres méthodes géophysiques.

International audience; Depuis le début de la pandémie de la COVID-19, de nombreux travaux ont cherché à modéliser et comparer les impacts des politiques d'intervention non pharmaceutique. Cette volonté s'est cependant heurtée à la difficulté à pouvoir facilement prendre en compte la variété des cultures, des comportements et des contextes dans lesquels se déploient à la fois le virus et les politiques de santé publique. Initié pour soutenir les autorités vietnamiennes dans le déploiement d'interventions à l'échelle locale, l'environnement logiciel COMOKIT, qui repose sur un modèle à base d'agents, a été conçu dès le départ pour être modulaire et facilement déployé sur différents cas d'étude, quelles que soient la disponibilité et la qualité des données démographiques, géographiques et épidémiologiques. Intégrant des outils de construction de l'environnement géographique et de génération de populations synthétiques, COMOKIT permet une modélisation détaillée des individus et de leurs activités à une temporalité fine. Cet article présente son adaptation à la comparaison de l'impact de mesures de confinement dans le centre-ville de Nice.

International audience; COVID-19 may become a norm in our day-to-day life the way flu did. However, given the unique characteristics of COVID-19 pandemic, our societies are not ready to effectively respond to extensive changes such as expensive lockdowns or teleworking. Society cannot afford operating, in the long run, under lockdown restrictions. Therefore, policies so far adopted by many politicians is to impose and relax restriction measures on periodic basis to assure the economy and wellbeing of the society are guaranteed while being able to manage the influx of hospitalized citizens. Nonetheless, the effectiveness of these restrictive policies, which are typically hypothetical, is to large extent unknown to us. Most of the already implemented policies are defined around quarantining people (including infected and uninfected). The main goal of these quarantine-based policies is to reduce the exposure of the susceptible population to the disease. However, several activities remain vital for the society, e.g., grocery shopping, and medical work trips. Thus, modeling human activities and their mobility patterns is critical for studying the spatial transmission of infectious diseases in large-scale urban areas. Activity-based models (ABMs) have been developed by transport modelers in the past decades replicating the mobility of people with high temporal and spatial resolution. We are working with such model (i.e., Albatross) which can simulate data-driven activity pattern and travel of all people in the Netherlands which such fine granularity. Albatross mimics the decision-making process of the agent based on decision-trees derived from activity diary data. We understand from this model who interacts with who, where, for how long, how is using which transport mode and so on. In addition to location and modes of transportation of activities, detailed information about demographic attributes of people can be provided by Albatross. In epidemiology, it is essential to know age, gender, and any health background of individuals. Nonetheless, the existing SEIR (Susceptible – Exposed – Infectious – Recovered) models are completely blind to these attributes of people because of their intrinsic crude aggregation in time, space and demographics. Against this background, we plan to combine the activity-based model (Albatross) with an epidemiology agent-based model (COMOKIT). COMOKIT-Albatross is a dedicated custom adaptation of COMOKIT. It simulates the spread of COVID-19 at the scale of the Netherlands. A large-scale (multi-million agents) simulation with the aid of Dutch supercomputer Cartesius. The activity diary for COMOKIT comes from the pre-generated mobility agenda in Albatross. The daily activities in Albatross are modelled at the precision of a minute, while the default COMOKIT behavior is modelled at the precision of one hour. The atomic spatial unit of this extension is at a Dutch 6-digit Postal Block level compared to the building level in COMOKIT. Epidemiological dynamics for COMOKIT-Albatross has also been modified to take into account this larger atomic spatial unit. The larger spatial resolution makes it possible to simulate a wider region, and fits the required spatial distribution of activities as modelled in Albatross agenda generator.

International audience; GAMA provides several tools to model the behavior of agents: reflexes, BDI architecture, systems of differential equations, etc. Nevertheless, in the last few years different plug-ins have been developed to enrich these possibilities. We present here 3 plug-ins: the argumentation plugin, the Bayesian network plugin and the artificial learning plugin. The first one allows modellers to use the Dung’s argumentation system to simulate opinion dynamics [1, 2]. More precisely, this plug-in allows to explicitly represent agents’ own mental deliberation process from arguments towards an opinion, through the use of the argumentation system of Dung. This plugin was already used to study the adoption of vegetarian diets and digital tools in agriculture. The second plugin allows modelers to build a bayesian network, i.e. a probabilistic graphical model that represents a set of variables and their conditional dependencies via a directed acyclic graph, and make decisions based on them: typically, knowing the probability of a variable from a set of other variables. The last plugin based on the open-source library Weka [3] allows the modeler to use a large number of supervised (CHAID, JRip, Multi-layer perceptrons, SMO, J48, Random Forest...) and unsupervised (K-means, EM,DB-scan, Cobweb...) learning algorithms. In particular, this plugin allows to build a classifier from a set of instances and to use the classifier to classify new instances.

International audience; GAMA is a modeling and simulation platform for creating simulations following a multi-agent approach. COMOKIT is a modeling kit written by GAMA for the purpose of analyzing and comparing the policies against the spread of the Covid 19 pandemic at the scale of a city. As a member of the COMOKIT project, with the mission of researching and developing to improve the user interaction and user interface on both GAMA and COMOKIT, we propose to develop a necessary feature for GAMA. That’s dashboard feature, a good means of communicating and analyzing information for the most fields today. Currently, GAMA platform provides users display chart types to support the analysis and visualization of data in the simulation, some of which have not yet available to put on a dashboard, for example: gauge type, metric type, data table type. However, the layout and the way to interact with the chart are limited, the output of simulation is not really user-friendly. This research will focus on creating a new syntax (GAML) in GAMA to use for displaying a dashboard in simulation. The generated syntax should be really easy to use, easy to understand and easy to edit. The dashboard is created allow users to arrange the charts depending on the research purpose. With the creation of a dashboard feature, it can be applied to specific research models such as COMOKIT, helping modelers to synthesize analytical data on the dashboard. From which to evaluate the results and effectivenesss of the research model correctly.

International audience; Agent-based model (ABM) are a kind of computer model that makes it possible to simulate a set of autonomous interacting programs called agents in a shared virtual environment. Among other application field, it has been commonly used to simulate social phenomena such as urban segregation, opinion dynamic or epidemiological crisis. Recently, a research emphasis has been put on ABM to study in silico the impact of non-pharmaceutical interventions to mitigate the SARS-CoV-2 outbreak of 2020, with few of them that had a great impact on global political responses. Among the model used COMOKIT has been design to simulate the every-day-life of inhabitant of various cities in Vietnam and test policy interventions for various COVID-19 spread scenarios. Such endeavor required huge computational power to handle a huge number of simulation replication over a large set of parameters. In this proposal we present a python package that enables to easily generate, explore and build reports for any COMOKIT experiment to be launched over High-Performance Computing (HPC) infrastructure.

International audience; Agent-based simulation has long been used to study the dynamics of innovation adoption and diffusion. However, the vast majority of this work is limited to a simplified representation of this process, which does not make it possible to explain the reasons for an agent’s change of opinion, an element that is nonetheless fundamental to understanding the dynamics of innovation diffusion. In order to overcome this limitation, we propose a generic agent-based model in which the knowledge of each agent is explicitly represented in the form of arguments, which carry information about the innovation. These arguments are the objects that the agents will exchange during their interactions. The advantage of this approach is that it allows to trace the state of knowledge of an agent in order to understand the evolution of its behavior in front of an innovation. We also propose to represent the decisional model of the agents with the Theory of Planned Behaviour (TPB). This theory, very classical in psychology, offers an integrative framework to formalize the behavior of agents. In TPB, the intention to behave is derived from three variables: the attitude, the subjective social norm and the perceived control of the behavior (PBC). The attitude represents the knowledge and opinion that an individual has about a behavior - in our case the adoption of the innovation. The subjective social norm is the individual’s perception of the adoption intention of his/her social network. Our proposal is to compute the attitude of the agents from their knowledge about the innovation, modeled as an argument graph, using the same approach as the one proposed by [2]. Concerning the social norm, we propose to be inspired by the work of [1], who suggests that during an interaction between two individuals the influence of one on the other depends on the opinions and the certainties that they have on the subject. At last, concerning PBC, which is specific to the type of innovation studied and to the individual concerned, we propose to transcribe it in the form of a variable specific to each individual, which may or may not be constant depending on the case of application. An application of this model is proposed to study the diffusion of communicating water meters by farmers on the Louts River (South-West of France).

International audience; In recent years, the combined effects of fluctuations in the benefit of agricultural and seafood production, management policies and environmental changes (increase of saltwater intrusion due to subsidence and sea level rise, increase of the length of drought periods because of the elevation of temperatures and the decrease of freshwater availability) have seriously affected the agricultural production of the coastal provinces in the Mekong Delta. Adaptation of farmers take many forms, including deep changes in their agricultural practices and land-use in order to maintain their income. However, the somewhat uncontrolled conversion of farming systems to follow market demand and get immediate profits will in some cases negatively impact the environment (degradation of soils, increase of salinization) and locally disrupt the long-term land-use plans established by provinces, normally designed to be sustainable at the global scale. In this work, we focus on land-use decision in the Mekong Delta. This region is planned to be a sustainable land-use types, but farmers tend to shift from rice to shrimp farming systems due to economic reasons. These decisions endanger the planned sustainable development of the area and the effects of infrastructures built to reach the plan. It is thus necessary to study and forecast this conversion to support the land-use planning and recommend the best suitable strategy to farmers. However, this forecasting work is currently lacking tools. It is mainly based on the human-made land inventory carried out every 5 years, which is obviously made under times and human resources limitations. This work aims to apply an Agent-based model to support managers' decision-making and to provide a picture of the future to encourage people to choose land-use types. The model is designed by agent-based approach where farmers are represented as the cell from the land-use map and the description of their states and behaviors. The simulation will let them interact and from these interactions, the dynamics of the system is expected to emerge. Decision-making process is based on the multicriteria analysis in which farmers will thus have to make their decisions given environmental, climate and socio-economic changes. The input data for verification and calibration of the model have been collected from the existed land-use map in 2015 and interpreted from Sentinel 2 satellite images in 2020 for the land-use. The climate change scenarios are forecasting to 2030. The results of the experiments showed evolutions of land-use in 2030 with climate change scenarios to analyze farming risks when the environment (including temperature, rainfall, water volume) and prices change in a harsher direction.

International audience; Natural disasters continue to cause tremendous damage to humans’ lives and properties. The Philippines, due to its geographic location, is considered a natural disaster-prone country experiencing an average of 20 tropical cyclones annually. This condition necessitates the need to study what can be done to mitigate the effects of weather-related disaster. Understanding what factors significantly affects decision making during crucial evacuation stages could help in making decisions on how to prepare for disasters, how to act appropriately and strategically respond during and after a calamity. In this work, an agentbased model of human behavior during typhoon evacuation is presented. In the model, civilians are represented by households and their evacuation decisions were based from their calculated perceived risk. Also, rescuer and shelter manager agents were included as facilitators during the preemptive evacuation process. National and municipal census data was employed for the model, particularly for the characteristics of household agents. Further, geospatial data of a village in a typhoon-susceptible municipality was used to represent the environment. In the model, household agents are placed randomly on buildings while rescuers roam the area. Shelter managers are in designated evacuation shelters and are stationary. Additionally, conceptualized formulas are provided to attain the individual evacuation decision of household agents. The decision to evacuate or not to evacuate depends on the agent’s perceived risk which also depends on three decision factors: characteristics of the decision maker (CDM); capacity related factors (CRF); and hazard related factors (HRF).Weights are assigned to each decision factor to assess their impact on the model outcome. Finally, the number of households who decided to evacuate or opted to stay as influenced by the model’s decision factors were determined during simulations. Linear regression was used to determine significant predictors to determine evacuation decision. Sensitivity analysis shows that all parameters used in the model are significant in the evacuation decision of household agents. Findings also show that total evacuation decision is more sensitive to weights assigned in capacity related factors, specifically type of house, floor level and past typhoon experience.

International audience; In recent years, the rapid development of infrastructure and means of transportation has required scientists to provide a systematic framework to analyze their impacts. Various works have been dedicated to the understand of vehicles movement patterns through cellular automaton models, with some notable works includes car-following model, lane-changing model, and gap acceptance model. However, while many agent-based models have been developed for cities' structured traffic patterns, many challenges still exist in modeling practical situations where different vehicle types do not always follow by lanes. Here, we introduce a new approach for mixed-traffic simulation using agent-based modeling, which comes with three novel developments: (1) naturalistic representation of stochastic collision at junctions, which could potentially be used to study the rate of traffic accidents in the mixed-lane traffic setting. (2) integration of Bezier curve into GAMA agent-based platform to improve the accuracy in modeling autonomous vehicles trajectories. (3) a pipeline to integrate our model with GIS data and real-time traffic information, which could be combined with other combinatoric optimization frameworks for logistics applications.

International audience; Since the appearance of Covid-19, many models have been developed to answer the question of what policies to implement to minimize its spread. Among these models, some are interested in the propagation inside buildings. Many works have indeed highlighted the impact of propagation in closed environments. It is in this context that we have developed the COMOKIT BUILDING model. This model is part of the COMOKIT framework (COVID-19 Modeling Kit) [1]. COMOKIT BUILDING is based on the model proposed by [2]. As in this model, the agents represent individuals who will come to do activities in the building. For that, each agent has an agenda that defines the activity it has to perform according to the time. They will then move in the building to go to their activity place and realize it. However, contrary to the model [2], which is primarily interested in the risk taken by individuals over a day, we are interested impact of the preventive measure over a longer period. Thus, we use the epidemiological model of COMOKIT (SEIR) to describe the evolution of the disease for individuals. In addition, taking inspiration from [2], we defined three types of transmission vectors (direct, airborne and via surfaces). Several types of interventions are taking into account: masks, physical distancing, room ventilation, physical barrier, etc. The model was implemented using GAMA 1.8.2 and takes advantage of some new features of this version, in particular the use the newly integrated pedestrian plugin for the agent movement. An application of the model to study the impact of prevention measures on the spread of the virus in a Hospital in the city of Danang in Vietnam will be presented. This hospital was at the origin of the formation of a cluster. Since then, several clusters have appeared in hospitals in Vietnam, making the issue of preventive measures in hospitals a major challenge for the Vietnamese government.

International audience; GAMA in 2021 is more or less 13 years old, depending on where we place the starting point of this great adventure. Heir of a series of modeling “platforms” (MANTA, EMF, …), first born as an extension of RePast to facilitate the handling of geographical data within agent-based models, influenced by great ancestors such as NetLogo, RePast, SWARM or CORMAS, made possible by the advent of IDEs such as Eclipse, the popularity of meta-modeling or the progresses in open-source code sharing, progressively provided with capabilities going well beyond the strict implementation of agents in agent-based models, used for applications that no one had foreseen ten years ago, GAMA does not have an exactly linear history ! While this (apparent) lack of roadmap has probably prevented a wider adoption of GAMA in the industrial (and even academic) world, it has allowed it to evolve with an amazing agility, and to be in particular flexible when incorporating other modeling approaches, open to new technologies and concepts, while remaining accessible to users and modelers. This presentation, hopefully avoiding making a boring summary, will go through some of the main stages of the short history of GAMA with the aim of triggering discussions and providing insights on how the platform can – and should -- evolve in the future: i.e, what mistakes not to make again? Which lessons can we draw from these 13 years in terms of agility and organization of the development of GAMA? Which forgotten development path could or should be promoted again?

International audience; Modeling traffic is about as old as car traffic itself. Among the existing modeling approaches, one of the most popular for traffic simulation is agent-based modeling. Agent-based models enable to describe each driver as an autonomous agent, making decisions based on its own aptitudes and local environment. With these models, each vehicle moves according to surrounding infrastructure and vehicles, accelerates and slows down according to its environment while maintaining a safe distance with other cars. These last few years have seen the number of traffic agent-based frameworks (MATSim, SUMO...) increase significantly. While these frameworks are appropriate for the study of normal traffic conditions, it is often complex to adapt them, in particular for non-computer scientists, to more specific application contexts such as the study of the impact of uncommon events (e.g. car accidents, evacuation following a technological disaster). In this context, we present the traffic plugin of the GAMA modeling and simulation platform, allowing the modeler to easily define new easily-tunable microscopic traffic simulations with a detailed representation of the driver’s operational behaviors. In particular, it allows the integration of road infrastructures and traffic signals, change of lanes by driver agents and their tendency to follow norms [2]. Moreover, the tool allows for city-level simulations with tens of thousands of driver agents. In addition, its latest version offers the possibility to define vehicles occupying different numbers of lanes allowing to simulate mixed traffic with cars, motorbikes, bus, etc. In addition, this version improves the speed management of vehicles by using the classic IDM (intelligent driver model) model. The plugin was validated and uses in many applications cases in France and in Vietnam, in particular to simulate city evacuation in disaster contexts [1].

International audience; In recent years, several agent-based simulation models have been produced to study complex systems in a wide range of domains. Three main stages are conducted: modelling, simulation and analysis of results, which represent the main focus of the majority of proposed agent-based simulation tools [1]. Participatory modelling aims at using modelling in support of a decision-making process involving stakeholders in one or several stages of the modelling process [2]. Participatory simulation aims at allowing stakeholders to interact with a simulated model at runtime, often through a serious game [1]. However, most of the existing agent-based simulation tools do not support the involvement of the stakeholders in the modelling and simulation process. In fact, this makes them unsuitable for situations that require real-time interaction with the stakeholders. For instance, for building serious games as a support for decision-making processes. In such situations, it is necessary to simplify interactions and integrate relevant feedbacks among complex systems into the model. So, the stakeholders can make direct changes within the simulation process, as well as see the results of their changes almost instantly. In light of the voids identified above, we aim to evolve GAMA to support participatory modelling. So, we developed a new tool, called GAMA Unity Client, that allows the visualization of simulations on different devices without having to install GAMA (e.g. devices running under android), and more importantly, allows stockholders to set up simulations and interact with the model during the simulation process. GAMA Unity Client consists of two parts: a GAMA plugin that implement a new agent’s skill (called unity) and a Unity Client, as a Unity generic scene that can be installed in different devices, with different operating systems. So, it relies on the GAMA platform; which offers several multi-agent modelling and simulation tools and ii) Unity; which is a cross-platform game engine. GAMA Unity Client relies on two existing tools in GAMA: i) The network skill, to manage network connections and messages exchanges and ii) The remote plugin which provides a reliable mechanism to publish (expose) data values to distant clients when changes occur during simulation. Here, the model is run on a distant GAMA platform, and the stockholders can interact with the model using the Unity Client on different devices connected to the network. The communication between the Unity Client and the GAMA Platform is made using the MQTT protocol.

International audience; Since its emergence in China late 2019, the COVID-19 pandemic has spread rapidly around the world. Faced with this unknown disease, public health authorities were forced to experiment, in a short period of time, with various combinations of interventions at different scales. The scientific community has quickly emerged as a centerpiece to resolve the various puzzles the virus was the origin of, including how to flatten the curve of infection, limit the spread of the pandemic, create and dispatch the vaccines, and what could be the treatments. One of the main expectations from policy makers was to be able to help them build an appropriate response and forecast the consequences: this has put lights upon epidemiological models like never before. To meet this demand, we have developed an agent-based modeling framework called COVID-19 Modeling Kit (COMOKIT), designed to be generic, scalable and thus portable in a variety of social and geographical contexts. COMOKIT combines models of person-to-person and environmental transmission, a model of individual epidemiological status evolution, an agenda-based 1-h time step model of human mobility, and an intervention model. It is designed to be modular and flexible enough to allow modelers and users to represent different strategies and study their impacts in multiple social, epidemiological or economic scenarios. Since its first version (COMOKIT 1.0 - presented in (Gaudou et al., 2020)), COMOKIT has deeply evolved, through different related projects such as the ANRS COMOKIT project, COMOKIT Camps on the transmission of covid in refugee camps, COMOKIT Azur on the impact of morpho-functional characteristics of cities or neighborhoods on the spread of the disease, COMOKIT Albatross on a large-scale application of COMOKIT in the Netherlands using data for the construction of individuals’ agendas: enrichment of the population generation process, better consideration of the link between the simulated territory and the outside area, accounting of variants and new types of interventions (e.g. vaccination), dashboard to better visualize data, multilevel modeling, etc. In this presentation, we propose to go back over COMOKIT, its evolutions, but also to look at the impact of COMOKIT on the development of the GAMA platform. Indeed, all the efforts put in COMOKIT have greatly shaped ideas for the future development of GAMA, mostly questioning the link between the platform, data and decision-making process.

International audience; According to recent studies, Vietnam is one of the twenty countries most affected by natural disasters in the world, and particularly by floods either on the low elevation coastal zones (risk of submersion) or along the Red River and the Mekong River (risk of flooding). In this context, dams are both means of mitigation but also threats given the possible failures and ruptures. The authorities must, therefore, prepare warning systems and evacuation plans for the downstream population to avoid loss of life. Agent-based models are now the approach of choice to support such preparedness by considering the system as a whole and integrating dynamics of different natures: hydrology, population behavior, evacuation, crisis management, etc. To design such a decision-support tool, modelers generally need to couple different formalisms, such as diffusion equations when considering the hydrodynamic part, and agent-based modeling when considering inhabitants’ behaviors. This is the goal of the ESCAPE project, which uses agent-based simulations to explore evacuation strategies and contribute to the development and evaluation of evacuation plans. In this study, to improve the ESCAPE framework, we propose to combine a hydraulic dam failure model with an agent-based evacuation model using the GAMA platform. We focus on the evacuation of a Hanoi city (Vietnam) district, threatened by flooding due to the failure of the Hoa Binh dam located more than 80 km upstream of the city. We demonstrate how to methodologically and operationally couple a hydrodynamic water diffusion model (implemented using the HEC-RAS software) and a multi-paradigm evacuation model (using the ESCAPE framework). Our goal is to extend and enrich this population evacuation model by coupling it with flood simulation.

International audience; The development of permanent or temporary pedestrian areas, whether for leisure or to decrease air pollution, has become an integral part of urban planning in numerous cities around the world. Hanoi, the capital of Vietnam, began to implement its first area, around the iconic Hoan Kiem lake, a few years ago. In most of cases, however, road closure is likely to deport traffic to nearby neighbourhoods with the consequences of intensifying congestion and, possibly, increasing air pollution in these areas. Because this outcome might appear counter-intuitive to most stakeholders, it is becoming more and more necessary to analyse, assess and share the impacts of these developments in terms of traffic and pollution shifts before implementing them. In this project, we used the GAMA platform to build an agent-based model that simulates the traffic, its emissions of air pollutants, and the diffusion of these pollutants in the district of Hoan Kiem. This simulation has been designed so as to serve either as a decision support tool for local authorities or as an awareness-raising tool for the general public: thanks to its display on a physical 3D model of the district, people can effectively and naturally interact with it at public venues. Although more accurate data and more realistic diffusion models are necessary and will need further research in the future, the simulation is already able to reflect traffic and air pollution peaks during rush hours, allowing residents and developers to understand the impact of pedestrianization on air quality in different scenarios.

International audience; Introduction : Are some urban environments more resistant than others to cope with a pandemic such as the one caused by COVID-19? This question is at the heart of many debates because this pandemic reminds us how vulnerable urban populations can be. It invites us to explore how policies could take advantage of the knowledge of local epidemic dynamics in given morpho-functional contexts to become more place-specific.

International audience; Agent-based simulation has long been used to study the dynamics of innovation adoption and diffusion. However, the vast majority of these works are limited to an abstract and simplified representation of this process, which does not make it possible to explain the reasons for an agent's change of opinion, an element that is nonetheless fundamental to understanding the dynamics of innovation diffusion. In order to overcome this limitation, we propose an agent-based model of innovation adoption and diffusion based on a classical theory of psychology, the theory of planned behaviour, and on formal argumentation. Each agent thus has the opportunity to exchange arguments with another agent and to build his/her opinion on an innovation from the set of arguments he/she knows. An application of this model is proposed to study the diffusion of communicating water meters by farmers on the Louts River (southwestern France).

ISBN: 978-604-9857-88-1 ; ISSN: 2525-2488; International audience; Dans le contexte de crues rapides en milieu urbain, le temps de réponse des populations est un facteur majeur de réduction des risques. L’intervalle entre la survenance de la crue et la mise en protection des populations est principalement impacté par le caractère rapide de l’aléa, par la structure du tissu urbain perturbant les dynamiques d’évacuation et par le contexte institutionnel de prévention et de mise en alerte des pays. Les approches de simulation à base d’agents visent à modéliser conjointement la dynamique de l’aléa avec celle des populations exposées pour proposer des solutions permettant de réduire les vulnérabilités individuelles lors de ces événements à dynamique rapide. La perspective proposée dans cette contribution est construite sur une analyse d’enquête de terrain, prolongée d’une modélisation des comportements de mise en protection et de leur simulation informatique à l’aide de systèmes multi-agents.

National audience; Cette contribution présente l'utilisation des réseaux de neurones convolutifs pour la détection d'anomalies magnétiques. L'approche développée permet la localisation de dipôles magnétiques, avec le comptage du nombre de dipôles, leur position géographique et la prédiction de leurs paramètres (moment magnétique, profondeur et déclinaison). Elle sera ensuite testée sur des données réelles, dans le cadre par exemple, d'une détection pyrotechnique pour la prospection de munitions non explosées, avant d'envisager une application vers d'autres méthodes géophysiques.

International audience; Ecohydrological processes are of primary importance in agricultural landscapes where fluxes of water are driven by vegetation, whether cultivated in fields or grasslands or present in semi-natural habitats and interstitial areas (open channels, riparian borders, inter-ranks of perennial crops) surrounding the cultivated parcels. Human practices are a key lever of the composition and properties of vegetation in landscape agroecosystems. In order to better understand their impact, we propose a suite of Gama and Matlab programs coupled in a single OPALE-gui (OPerational Assessment of Landscapes Eco-functionalities) that aims at (i) representing landscape organization in relation with the functioning and decision rules of annual crops or breeding systems, (ii) evaluating water movements and trajectories from biomass and hydrological exchanges, and (iii) analyzing the transfer dynamics of nutrients, suspended matter or fecal bacteria based on particle tracking methods. More precisely, the tool includes a set of libraries about: • the distribution of the farmland part of the landscape, allowing to spatially allocate crops into farm fields and the related farming practices (inputs, soil tillage, fodder management...) following time schedules (Land Use and Land Cover Change model, LULCC); • the modeling of water movements within previous simulated landscapes, considering surface, sub-surface and deep flows (Water Movements Within Landscapes model, WMWL); • the evaluation of –inert and living– solute and suspended matter transfers related to water movements, including N, P, E. Coli (EC) and Suspended Matter (SM); • the construction of output normalized indicators of landscape function in order to assess its regulation service. OPALE allows to better understand the underlying processes that drive exchanges and sink-source effects operating at the landscape scale, including a landscape transfer function and a delivery ratio for the global assessment of landscapes. It was tested in three agricultural contexts and biophysical situations. Water flows and flows of associated matters were compared to data recorded at the catchment outlet, and showed the efficiency of the algorithms developed in the generic OPALE libraries.

Open Code; International audience; This paper introduces a generic agent-based model simulating the exchange and the diffusion of pro and con arguments. It is applied to the case of the diffusion of vegetarian diets in the context of a potential emergence of a second nutrition transition. To this day, agent-based simulation has been extensively used to study opinion dynamics. However, the vast majority of existing models have been limited to extremely abstract and simplified representations of the diffusion process. These simplifications impairs the realism of the simulations and disables the understanding of the reasons for the shift of an actor's opinion. The generic model presented here explicitly represents exchanges of arguments between actors in the context of an opinion dynamic model. In particular, the inner attitude towards an opinion of each agent is formalized as an argumentation graph and each agent can share arguments with other agents. Simulation experiments show that introducing attacks between arguments and a limitation of the number of arguments mobilized by agents has a strong impact on the evolution of the agents' opinion. We also highlight that when a new argument is introduced into the system, the quantity and the profile of the agents receiving the new argument will impact the evolution of the overall opinion. Finally, the application of this model to vegetarian diet adoption seems consistent with historical food behaviour dynamics observed during crises.

International audience; Leveraging artificial intelligence (AI) approaches in animal health (AH) makes it possible to address highly complex issues such as those encountered in quantitative and predictive epidemiology, animal/human precision-based medicine, or to study host × pathogen interactions. AI may contribute (i) to diagnosis and disease case detection, (ii) to more reliable predictions and reduced errors, (iii) to representing more realistically complex biological systems and rendering computing codes more readable to non-computer scientists, (iv) to speeding-up decisions and improving accuracy in risk analyses, and (v) to better targeted interventions and anticipated negative effects. In turn, challenges in AH may stimulate AI research due to specificity of AH systems, data, constraints, and analytical objectives. Based on a literature review of scientific papers at the interface between AI and AH covering the period 2009–2019, and interviews with French researchers positioned at this interface, the present study explains the main AH areas where various AI approaches are currently mobilised, how it may contribute to renew AH research issues and remove methodological or conceptual barriers. After presenting the possible obstacles and levers, we propose several recommendations to better grasp the challenge represented by the AH/AI interface. With the development of several recent concepts promoting a global and multisectoral perspective in the field of health, AI should contribute to defract the different disciplines in AH towards more transversal and integrative research.

International audience; Irrigation systems contribute worldwide to the provision of a wide range of services on which the survival and well-being of humanity depend. They are of primary importance in Vietnam where about 90% of the water used is for irrigation and aquaculture and where agriculture is the largest employer and a major contributor to the national GDP and to the income of the low-salary households. Nevertheless, irrigation systems have recently been subjected to several issues including increasing demand, pollution, under-investment, depletion of resources or environmental changes. Any mitigation measure against these issues needs to be sustainable with respect to the very diverse uses of the water, the changing conditions upstream and downstream, and the somewhat conflicting objectives carried out by land-use/agricultural planning on one hand and urbanization and society well-being on the other. This need of sustainability requires the design of innovative tools to tackle these issues. This work aims at exploring the usage of Agent-Based Modelling coupled with a tangible and interactive interface in order to enhance interactions between stakeholders and support the evaluation of various alternatives of the management of the Bac Hung Hai irrigation system.

International audience; Pedestrian zones are present in numerous cities around the world, and Hanoi city began to organize one a few years ago. However, closing roads can lead to heavy traffic congestion in surrounding areas and, consequently, more air pollution in these areas. There is, therefore, a need for analyzing and predicting the outcomes in terms of air pollution when certain roads are closed, before actually implementing a plan. In this project, we used the GAMA platform to build an agent-based model that simulates the traffic and air quality in Hoan Kiem district. This model can be used as a decision support tool for local authorities and as an information tool for the general public: thanks to its output on a tangible interface, people can interact with the simulation at public venues and explore various scenarios. Although more accurate data and realistic diffusion models are still lacking and will need further research in the future, the simulation is alreay able to reflect traffic and air pollution peaks during rush hours quite realistically.

The Internet contains a wealth of information in the form of unstructured texts such as customer comments on products, events and more. By extracting and analyzing the opinions expressed in customer comments in detail, it is possible to obtain valuable opportunities and information for customers and companies. The model proposed by Jebbara and Cimiano. for the extraction of aspects, winner of the SemEval2016 competition, suffers from the absence of lexico-grammatic input characteristics and poor performance in the detection of compound aspects. We propose the model based on a recurrent neural network for the task of extracting aspects of an entity for sentiment analysis. The proposed model is an improvement of the Jebbara and Cimiano model. The modification consists in adding a CRF to take into account the dependencies between labels and we have extended the characteristics space by adding grammatical level characteristics and lexical level characteristics. Experiments on the two SemEval2016 data sets tested our approach and showed an improvement in the F-score measurement of about 3.5%., L'analyse des opinions consiste à extraire des connaissances à partir des commentaires laissés par les utilisateurs à propos d'un produit, service, texte,... L'analyse des opinions basée sur les aspects consiste alors à décomposer le commentaire afin d'extraire les aspects que cet utilisateur a évalué. Le modèle proposé par Jebbara et Cimiano, vainqueur de la compétition SemEval2016 n'extrait pas correctement des aspects composés et ne prend pas en compte les caractéristiques lexico-grammaticales des textes en entrée, ce qui limite aussi ses performances dans la détection des aspects. Nous proposons une amélioration du modèle de Jebbara et Cimiano. en y introduisant des unités CRF afin de prendre en compte les dépendances entre les étiquettes et ajoutant aux entrées du modèle des caractéristiques lexico-grammaticales. Les expérimentations faites sur les deux jeux de données de SemEval2016 ont permis de tester cette approche et montrer une amélioration de la mesure F-score d'environ 3.5%. ABSTRACT. The Internet contains a wealth of information in the form of unstructured texts such as customer comments on products, events and more. By extracting and analyzing the opinions expressed in customer comments in detail, it is possible to obtain valuable opportunities and information for customers and companies. The model proposed by Jebbara and Cimiano. for the extraction of aspects, winner of the SemEval2016 competition, suffers from the absence of lexico-grammatic input characteristics and poor performance in the detection of compound aspects. We propose the model based on a recurrent neural network for the task of extracting aspects of an entity for sentiment analysis. The proposed model is an improvement of the Jebbara and Cimiano model. The modification consists in adding a CRF to take into account the dependencies between labels and we have extended the characteristics space by adding grammatical level characteristics and lexical level characteristics. Experiments on the two SemEval2016 data sets tested our approach and showed an improvement in the F-score measurement of about 3.5%.

International audience; Since its emergence in China, the COVID-19 pandemic has spread rapidly around the world. Faced with this unknown disease, public health authorities were forced to experiment, in a short period of time, with various combinations of interventions at different scales. However, as the pandemic progresses, there is an urgent need for tools and methodologies to quickly analyze the effectiveness of responses against COVID-19 in different communities and contexts. In this perspective, computer modeling appears to be an invaluable lever as it allows for the in silico exploration of a range of intervention strategies prior to the potential field implementation phase. More specifically, we argue that, in order to take into account important dimensions of policy actions, such as the heterogeneity of the individual response or the spatial aspect of containment strategies, the branch of computer modeling known as agent-based modeling is of immense interest. We present in this paper an agent-based modeling framework called COVID-19 Modeling Kit (COMOKIT), designed to be generic, scalable and thus portable in a variety of social and geographical contexts. COMOKIT combines models of person-to-person and environmental transmission, a model of individual epidemiological status evolution, an agenda-based 1-h time step model of human mobility, and an intervention model. It is designed to be modular and flexible enough to allow modelers and users to represent different strategies and study their impacts in multiple social, epidemiological or economic scenarios. Several large-scale experiments are analyzed in this paper and allow us to show the potentialities of COMOKIT in terms of analysis and comparison of the impacts of public health policies in a realistic case study.