Literatura académica sobre el tema "Heterogenous domain"

In this dissertation, we study models for coupled active--passive pedestrian dynamics from mathematical analysis and simulation perspectives. The general aim is to contribute to a better understanding of complex pedestrian flows. This work comes in three main parts, in which we adopt distinct perspectives and conceptually different tools from lattice gas models, partial differential equations, and stochastic differential equations, respectively. In part one, we introduce two lattice models for active--passive pedestrian dynamics. In a first model, using descriptions based on the simple exclusion process, we study the dynamics of pedestrian escape from an obscure room in a lattice domain with two species of particles (pedestrians). The main observable is the evacuation time as a function of the parameters caracterizing the motion of the active pedestrians. Our Monte Carlo simulation results show that the presence of the active pedestrians can favor the evacuation of the passive ones. We interpret this phenomenon as a discrete space counterpart of the so-called drafting effect. In a second model, we consider again a microscopic approach based on a modification of the simple exclusion process formulated for active--passive populations of interacting pedestrians. The model describes a scenario where pedestrians are walking in a built environment and enter a room from two opposite sides. For such counterflow situation, we have found out that the motion of active particles improves the outgoing current of the passive particles. In part two, we study a fluid-like driven system modeling active--passive pedestrian dynamics in a heterogenous domain. We prove the well-posedness of a nonlinear coupled parabolic system that models the evolution of the complex pedestrian flow by using special energy estimates, a Schauder's fixed point argument and the properties of the nonlinearity's structure. In the third part, we describe via a coupled nonlinear system of Skorohod-like stochastic differential equations the dynamics of active--passive pedestrians dynamics through a heterogenous domain in the presence of fire and smoke. We prove the existence and uniqueness of strong solutions to our model when reflecting boundary conditions are imposed on the boundaries. To achieve this we used compactness methods and the Skorohod's representation of solutions to SDEs posed in bounded domains. Furthermore, we study an homogenization setting for a toy model (a semi-linear elliptic equation) where later on our pedestrian models can be studied.

With the growing amount of data generated in large heterogeneous repositories (such as the Word Wide Web, corporate repositories, citation databases), there is an increased need for the end users to locate relevant information efficiently. Text Classification (TC) techniques provide automated means for classifying fragments of text (phrases, paragraphs or documents) into predefined semantic types, allowing an efficient way for organising and analysing such large document collections. Current approaches to TC rely on supervised learning, which perform well on the domains on which the TC system is built, but tend to adapt poorly to different domains. This thesis presents a body of work for exploring adaptive TC techniques across hetero- geneous corpora in large repositories with the goal of finding novel ways of bridging the gap across domains. The proposed approaches rely on the exploitation of domain knowledge for the derivation of stable cross-domain features. This thesis also investigates novel ways of estimating the performance of a TC classifier, by means of domain similarity measures. For this purpose, two novel knowledge-based similarity measures are proposed that capture the usefulness of the selected cross-domain features for cross-domain TC. The evaluation of these approaches and measures is presented on real world datasets against various strong baseline methods and content-based measures used in transfer learning. This thesis explores how domain knowledge can be used to enhance the representation of documents to address the lexical gap across the domains. Given that the effectiveness of a text classifier largely depends on the availability of annotated data, this thesis explores techniques which can leverage data from social knowledge sources (such as DBpedia and Freebase). Techniques are further presented, which explore the feasibility of exploiting different semantic graph structures from knowledge sources in order to create novel cross- domain features and domain similarity metrics. The methodologies presented provide a novel representation of documents, and exploit four wide coverage knowledge sources: DBpedia, Freebase, SNOMED-CT and MeSH. The contribution of this thesis demonstrates the feasibility of exploiting domain knowl- edge for adaptive TC and domain similarity, providing an enhanced representation of docu- ments with semantic information about entities, that can indeed reduce the lexical differences between domains.

A SEMantic Integration System (SemIS) allows a query over one database to be answered using the knowledge managed in multiple databases in the system. It does so by translating a query across the collaborative databases in which data is autonomously managed in heterogeneous schemas. In this thesis, we investigate the challenges that arise in enabling domain heterogeneous (DH) databases to collaborate in a SemIS. In such a setting, distributed databases modeled as independent data sources are pairwise mapped to form the semantic overlay network (SON) of the SemIS. We study two problems we believe are foremost to allow a SemIS to integrate DH data sources. The first problem tackled in this thesis is to efficiently organize data sources so that query answering is efficient despite the increased level of source heterogeneity. This problem is modeled as an “Acquaintance Selection” problem and our solution helps data sources to choose appropriate acquaintances to create schema mappings with and therefore allows a SemIS to have a single-layered and flexible SON. The second problem tackled in this thesis is to allow aggregate queries to be translated across domain heterogeneous (DH) data sources where objects are usually represented and managed at different granularity. We focus our study on relational databases and propose novel techniques that allow a (non-aggregate) query to be answered by aggregations over objects at a finer granularity. The new query answering framework, named “decomposition aggregation query (DAQ)” processing, integrates data sources holding information in different domains and different granularity. New challenges are identified and tackled in a systematic way. We studied query optimizations for DAQ to provide efficient and scalable query processing. The solutions for both problems are evaluated empirically using real-life data and synthetic data sets. The empirical studies verified our theoretical claims and showed the feasibility, applicability (for real-life applications) and scalability of the techniques and solutions.

Des milliers d'œuvres musicales sont décrites dans des catalogues des institutions culturelles, dont le rôle est de stocker toutes les créations musicales à travers le catalogage et de les diffuser auprès du grand public. Cette thèse s’inscrit dans le cadre du projet ANR DOREMUS -DOnnées en REutilisation pour la Musique en fonction des USages- qui vise à explorer les métadonnées des catalogues de trois grandes institutions culturelles : Bibliothèque Nationale de France (BNF), Philharmonie de Paris et Radio France afin qu'elles puissent communiquer entre elles et être mieux utilisées par les différents publics. Dans cette thèse, nous nous intéressons aux liens dits d’identité, exprimant une équivalence entre deux ressources différentes décrivant la même entité du monde réel. Notre objectif principal est de proposer une approche de liage générique, traitant certains challenges, avec comme cas concret d’utilisation les données de DOREMUS.Dans cette thèse, nous nous focalisons sur trois principaux challenges : (1) réduire la configuration manuelle de l’outil de liage, (2) faire face à différents types d’hétérogénéité entre les descriptions, et (3) Supprimer l’ambiguïté entre les ressources très similaires dans leur descriptions mais qui ne sont pas équivalentes. Certaines approches de liage demandent souvent l’intervention de l’utilisateur pour configurer certains paramètres. Ceci peut s’avérer être une tâche coûteuse pour l’utilisateur qui peut ne pas être expert du domaine. Par conséquent, une des questions de recherche que nous nous posons est comment réduire autant que possible l’intervention humaine dans le processus de liage des données. De plus, les descriptions des ressources peuvent présenter diverses hétérogénéités qu’un outil doit savoir gérer. Par ailleurs, les descriptions peuvent être exprimées dans différentes langues naturelles, avec des vocabulaires différents ou encore avec des valeurs différentes. La comparaison peut alors s’avérer très difficile en raison des variations selon trois dimensions : basées sur les valeurs, ontologiques et logiques. Dans cette thèse, nous analysons les aspects d’hétérogénéité les plus récurrents en identifiant un ensemble de techniques qui peuvent leur être appliquées. Un autre défi est la distinction entre des descriptions de ressources fortement similaires mais non équivalentes. En leur présence, la plupart des outils existants se voient diminuer leur efficacité en terme de qualité, en générant beaucoup de faux positifs. Dans cette optique, certaines approches ont été proposées pour identifier un ensemble de propriétés discriminatives appelées des clefs. De telles approches découvrent un très grand nombre de clés. La question qui se pose est de savoir si toutes les clés permettent de découvrir les mêmes paires d’instances équivalentes, ou si certaines sont plus significatives que d'autres. Aucune approche ne fournit de stratégie pour classer les clefs générées en fonction de leur efficacité à découvrir les bons liens. Afin d’assurer des alignements de qualité, nous avons proposé dans ce travail une nouvelle approche de liage de données visant à relever les défis décrits ci-dessus.Un outil de liage automatique de données hétérogènes, nommé Legato, qui répond aux challenges évoqués précédemment a été développé. Il est basé sur la notion de profile d’instance représentant chaque ressource comme un document textuel de littéraux gérant une variété d’hétérogénéités de données sans l’intervention de l’utilisateur. Legato implémente également une étape de filtrage de propriétés dites problématiques permettant de nettoyer les données du bruit susceptible de rendre la tâche de comparaison difficile. Pour pallier au problème de distinction entre les ressources similaires dans leur description, Legato implémente un algorithme basé sur la sélection et le ranking des clefs afin d’améliorer considérablement la précision au niveau des liens générés<br>This thesis is part of the ANR DOREMUS project. We are interested in the catalogs of three cultural institutions: BNF (Bibliothèque Nationale de France), Philharmonie de Paris and Radio France, containing detailed descriptions about music works. These institutions have adopted the Semantic Web technologies with the aim of making these data accessible to all and linked.The links creation becomes particularly difficult considering the high heterogeneity between the descriptions of the same entity. In this thesis, our main objective is to propose a generic data linking approach, dealing with certain challenges, for a concrete application on DOREMUS data. We focus on three major challenges: (1) reducing the tool configuration effort, (2) coping with different kinds of data heterogeneities across datasets and (3) dealing with datasets containing blocks of highly similar instances. Some of the existing linking approaches often require the user intervention during the linking process to configure some parameters. This may be a costly task for theuser who may not be an expert in the domain. Therefore, one of the researchquestions that arises is how to reduce human intervention as much as possible inthe process of data linking. Moreover, the data can show various heterogeneitiesthat a linking tool has to deal with. The descriptions can be expressed in differentnatural languages, with different vocabularies or with different values. The comparison can be complicated due to the variations according to three dimensions: value-based, ontological and logical. Another challenge is the distinction between highly similar but not equivalent resource descriptions. In their presence, most of the existing tools are reduced in efficiency generating false positive matches. In this perspective, some approaches have been proposed to identify a set of discriminative properties called keys. Very often, such approaches discover a very large number of keys. The question that arises is whether all keys can discover the same pairs of equivalent instances, or ifsome are more meaningful than others. No approach provides a strategy to classify the keys generated according to their effectiveness to discover the correct links.We developed Legato — a generic tool for automatic heterogeneous data linking.It is based on instance profiling to represent each resource as a textual documentof literals dealing with a variety of data heterogeneities. It implementsa filtering step of so-called problematic properties allowing to clean the data ofthe noise likely to make the comparison task difficult. To address the problem ofsimilar but distinct resources, Legato implements a key ranking algorithm calledRANKey

Afin de répondre aux besoins de l'industrie pétrolière d'une description plus fine de la géométrie et des propriétés pétrophysiques des bassins et des réservoirs, la simulation numérique des écoulements en milieux poreux doit évoluer vers des algorithmes plus performants et plus robustes vis à vis de la taille des simulations, de la complexité des maillages et des hétérogénéités du milieu poreux. Les méthodes de décomposition de domaine constituent une alternative aux méthodes multigrilles et pourraient permettre de lever les difficultés précédentes en terme de robustesse et d'efficacité sur architectures parallèles. Elles sont par nature plus adaptées au calcul parallèle et sont plus robustes en particulier lorsque les sous domaines sont résolus par des méthodes directes. Elles permettent aussi de traiter dans un cadre unique les couplages de modèles comme les puits ou les failles conductrices et s'étendent au cas des systèmes couplés. Le travail de thèse traite plus particulièrement de méthodes définies au niveau algébrique. On ne suppose pas avoir une connaissance préalable du problème continu dont la matrice provient. On n'a pas non plus accés aux matrices avant assemblage. Ce manque d'informations a priori rend plus difficile la construction de méthodes efficaces. On propose deux nouvelles méthodes de construction de méthodes de décomposition de domaine au niveau algébrique: la construction de conditions d'interface optimisées et d'une grille grossière. Ce dernier point est particulièrement important pour avoir des méthodes robustes vis à vis du nombre des sous-domaines. Les méthodes sont adaptatives et basées sur l'analyse de l'espace de Krylov généré durant les premières itérations de la méthode de Schwarz classique. A partir des vecteurs de Ritz correspondant aux plus basses valeurs propres, on construit des conditions d'interface et des grilles grossières qui annihilent l'erreur sur ces composantes. Les méthodes ont été testées sur des calculateurs parallèles pour des matrices issues de la simulation de milieux poreux.

To increase market penetration, mobile software makers support their popular applications on all major software platforms, which currently include Android, iOS, and Windows Phone. Although these platforms often offer a drastically different look and feel, cross-platform applications deliver the same core functionality to the end user. Maintaining and evolving such applications currently requires replicating all the changes across all supported variants, a laborious and intellectually taxing enterprise. The state-of-the-practice automated source translation tools fall short, as they are incapable of handling the structural and idiomatic differences of the software frameworks driving major mobile platforms. In addition, popular mobile applications increasingly make use of distributed resources. Certain domains, including social networking, productivity enhancement, and gaming, require different application instances to continuously exchange information with each other. The current state of the art in supporting communication across heterogeneous mobile devices requires the programmer to write platform-specific, low-level API calls that are hard not only to develop but also to evolve and maintain. This thesis reports on the findings of two complementary research activities, conducted with the goal of facilitating the development and communication across heterogeneous mobile devices: (1) a programming model and runtime support for heterogeneous device-to-device communication across mobile applications; (2) a source code recommendation system that synthesizes code snippets from web-based programming resources, based on the functionality written for Android or iOS and vice versa. The conceptual and practical advancements of this research have potential to benefit fellow researchers as well as mobile software developers and users.<br>Master of Science

Reactive transport models are essential tools for predicting contaminant fate and transport in the subsurface and for designing effective remediation strategies. Sound understanding of subsurface mixing in heterogeneous porous media is the key for the realistic modeling of reactive transport. This dissertation aims to investigate the extent of mixing and improve upscaling effective macroscopic models for mixing-controlled reactive transport in connected heterogeneous formations, which usually exhibit strongly anomalous transport behavior. In this research, a novel approach is developed for an accurate geostatistical characterization of connected heterogeneous formations transformed from Gaussian random fields. Numerical experiments are conducted in such heterogeneous fields with different connectivity to investigate the performance of macroscopic mean transport models for simulating mixing-controlled reactive transport. Results show that good characterization of anomalous transport of a conservative tracer does not necessarily mean that the models may characterize mixing well and that, consequently, it is questionable that the models capable of characterizing anomalous transport behavior of a conservative tracer are appropriate for simulating mixing-controlled reactive transport. In connected heterogeneous fields with large hydraulic conductivity variances, macroscopic mean models ignoring concentration variations yield good prediction, while in fields with intermediate conductivity variances, the models must consider both the mean concentration and concentration variations, which are very difficult to evaluate both theoretically and experimentally. An innovative and practical approach is developed by combining mean conservative and reactive breakthrough curves for estimating concentration variations, which can be subsequently used by variance transport models for prediction. Furthermore, a new macroscopic framework based on the dual-permeability conceptualization is developed for describing both mean and concentration variation for mixing-controlled reactive transport. The developed approach and models are validated by numerical and laboratory visualization experiments. In particular, the new dual-permeability model demonstrates significant improvement for simulating mixing-controlled reactive transport in heterogeneous media with intermediate conductivity variances. Overall, results, approaches and models from this dissertation advance the understanding of subsurface mixing in anomalous transport and significantly improve the predictive ability for modeling mixing-controlled reactive transport in connected heterogeneous media.