In this paper we view learning as an unconstrained non-linear minimization problem in which the objective function is defined by the negative log-likelihood function and the search space by the parameter space of an origin constrained product unit neural spatial interaction model. We consider Alopex based global search, as opposed to local search based upon backpropagation of gradient descents, each in combination with the bootstrapping pairs approach to solve the maximum likelihood learning problem. Interregional telecommunication traffic flow data from Austria are used as test bed for comparing the performance of the two learning procedures. The study illustrates the superiority of Alopex based global search, measured in terms of Kullback and Leibler's information criterion.

No description available.

L’affinage technique et l’adoption récente des technologies tactiles multi-points par les industriels et les utilisateurs ont fixé l’attention des designers d’interaction sur ces technologies. Tandis que de nombreuses études en interaction homme-machine se sont intéressées à comparer la performance de ces interfaces à celle des interfaces WIMP traditionnelles, peu se sont attachées à intégrer dans leur approche les spécificités du canal gestuel et les modalités d’interactions multi-tactiles. Dans cette étude, je défends l’idée que le design de telles interactions ne peut être approché qu’en suivant un modèle d’interaction spécifique intégrant l’ensemble des composantes physiques, cognitives, sensorielles et motrices du geste dans le couplage homme-machine. J’articule ma recherche autour d’un espace de conception, courte analyse sociotechnique de mon objet d’étude, dans lequel je définis un modèle d’interaction descriptif et génératif. Je détermine un ensemble de principes conceptuels et techniques permettant l’évaluation et la conception du design d’interfaces multi-tactiles de manière systémique et extensible. Au cours de cette étude, je précise les limites du paradigme d’ « interface naturelle » en nuançant les effets du réalisme des interactions dans l’efficacité de tels systèmes. Enfin, je présente les travaux de conception et de développement d’un environnement de développement réalisé dans le cadre d’un dispositif CIFRE qui a accompagné cette étude et permis d’enrichir le modèle théorique. / The technical refinement and the recent adoption of multi-touch technologies by both the industry and users made these technologies a major concern for interaction designers. While many studies on human-computer interaction have started to compare this type of interface with traditional WIMP interfaces performance-wise, few have included in their approach the gestural and system mapping specificities of multi-touch interactions. In this study, I defend the idea that such interaction design should follow a specific interaction model taking into account physical, cognitive, sensitive and motor aspects of gestures in the human-computer relation. I define a design space, sort of socio-technical approach, which participates in the definition of a descriptive and generative interaction model. I establish a set of conceptual and technical principles that allow the evaluation and conception of multi-touch interfaces, in a systemic and extensible way. Throughout this study, I emphasize the limits of the « natural user interface » paradigm by having a better understanding of how interaction realism affects system efficiency. Finally, I present a multi-touch framework developed as part of a CIFRE collaboration, which helped this study in the design and the extension of the conceptual model.

京都大学 / 0048 / 新制・課程博士 / 博士(工学) / 甲第18256号 / 工博第3848号 / 新制||工||1590(附属図書館) / 31114 / 京都大学大学院工学研究科都市社会工学専攻 / (主査)教授 小林 潔司, 教授 谷口 栄一, 准教授 松島 格也 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DFAM

<p>Utilizing a genetic algorithm training of the atom dipole interaction model was performed to arrive at C,H, N, & O atomic polarizabilities that constitute a correction to semiempirical molecular polarizability calculations increasing the accuracy of these calculations to near parity with DFT at a fraction of the computational load.</p>

Indiana University-Purdue University Indianapolis (IUPUI) / Molecular polarizability plays a significant role in chemistry, biology, and medicine. Classical prediction of polarizability often relies on atomic-type specific polarizability optimized for training set molecules, which limits the calculations to systems of similar chemical environment. Although ab initio (AI) quantum mechanical (QM) methods are more transferable in predicting molecular polarizability, their high computational costs especially when used with large basis sets for obtaining quantitatively reliable results make them less practical. To obtain accurate QM polarizability in an efficient manner, we have developed a dual-level approach, where the polarizability (α) obtained from the efficient semiempirical QM (SE) method is corrected using a set of element-base atomic polarizabilities derived from the atomic dipole interaction model (ADIM) to reproduce the density functional theory (DFT) results. We have optimized the atomic polarizability correction parameters for CHON-containing systems using a small training set of molecules and tested the resulting SE-ADIM model on the neutral drug-like molecules in the QM7B database. SE-ADIM corrected AM1 showed substantial improvement with its relative percent error (RPE) compared to B3LYP reduced from 33.81% to 3.35%. To further test its robustness for larger molecules in broader chemical bonding situations, we applied this method to a collection of drug molecules from the e-Drug3D database. For the 1004 molecules tested, our SE-ADIM model, which only contains four empirical parameters, greatly reduces the RPE in AM1 polarizability relative to B3LYP from 26.8% to 2.9%. Error decomposition shows consistent improvements across molecules with diverse bond saturations, molecular sizes, and charge states. In addition, we have applied AlphaML, a promising machine learning (ML) technique for predicting molecular polarizability, to the e-Drug3D dataset to compare its performance with our SE-ADIM correction of AM1. We found SE-ADIM performs competitively with AlphaML bolstering our confidence in the value of our method. Errors distinct to AlphaML were also discovered. We found four molecules for which AlphaML predicts negative molecular polarizabilities, all of which were peroxides. In contrast, SE-ADIM has no such issue with these molecules or this chemical type. Finally, to improve performance of SE-ADIM when correcting AM1 molecular polarizability calculations for charged molecules, we introduce a charge dependent polarizability (CDP) enabled SE-ADIM. Training the CDP enabled SE-ADIM with a single additional parameter, B, we were able to reduce error in AM1 molecular polarizability calculations of charged molecules relative to B3LYP from 29.57% to 5.16%. By contrast, SE-ADIM without CDP corrected AM1 relative to B3LYP had an RPE of 8.56%. The most benefit of CDP was evident within negatively charged molecules where AM1 error relative to B3LYP fell from 32.20% to 3.77% while SE-ADIM without CDP enabled error for these same negative molecules was 10.06%.

Storicamente, lo studio delle Nuove Tecnologie ha visto l’evoluzione di diverse discipline interessate alla valutazione delle stesse, in termini di sicurezza, usabilità, esperienza globale (dall’Ergonomia fino alla User Experience, passando per la cyberpsicologia). Tali discipline hanno identificato nei “bisogni degli utenti” la guida fondamentale di valutazione e design. Tuttavia, è mancata una contestualizzazione teorica del concetto di bisogno dell’utente. A tal proposito, questa tesi propone un modello teorico (PIM) basato sulla relazione tra tecnologie e intenzioni degli utenti. Queste ultime, sulla base della letteratura psicologica e filosofica, sono viste come l’antecedente mentale delle azioni, strutturate gerarchicamente. Sulla base del modello, vengono presentati tre esperimenti incentrati sull’interazione con le tecnologie e le intenzioni al livello motorio, prossimale e distale. I risultati del primo esperimento indicano che l’utilizzo di tecnologie interattive può modificare il contenuto rappresentazionale delle intenzioni motorie. Il secondo esperimento mostra che la presenza di intenzioni prossimali in utenti orienta la loro percezione di opportunità d’azione e la valutazione della tecnologia. Il terzo esperimento dimostra che la presenza di intenzioni distali negli utenti influenza la valutazione di usabilità di una tecnologia e le emozioni ad essa legate. La conclusione della tesi approfondisce l’importanza dell’analisi delle intenzioni nella valutazione delle tecnologie. / Historically, the study of New Technologies has seen the evolution of the various disciplines involved in technology evaluation, in terms of safety, usability, and the overall experience (from ergonomics to User Experience, and cyberpsychology). These disciplines have identified the "user needs" as the fundamental guideline for evaluation and design. However, a theoretical contextualization of the concept of user needs is still missing. This dissertation proposes a theoretical model (PIM) based on the relationship between technology and user intentions. Based on the psychological and philosophical literature, it defines intentions as the hierarchically-structured mental antecedents of actions. Starting from the model, three experiments are presented focusing on the interaction with the technologies and motor, proximal and distal intentions. The results of the first experiment indicate that the use of interactive technologies can modify the representational content of motor intentions. The second experiment shows that the presence of proximal intentions in users drives their perception of affordances in the technologies and their final evaluation. The third experiment shows that users’ distal intentions influence the evaluation of usability in the technology and the experience of usage-related emotions. The conclusion explores the importance of intention analysis in technology evaluation.

人口流動具有各種型態。其中包含了遷移、移動、以及通勤人口。在宏觀模型框架下，空間互動模型(簡稱SIM)對於測量人口流動扮演了重要的角色。距離遞減效應為空間互動模型中重要的因子。該效應描述了人口流動的頻率會隨著移動距離而逐漸下降。然而，從實證上，本研究發現人口流動與移動距離的函數，並非在距離上保有恆定的關係。 在本文中，我們提出了對此非恆定的距離遞減效應之修正方法。本修正法運用了轉折點模型的特點，套入了空間互動模型的距離函數上。本文首先運用了電腦模擬驗證了此方法的穩定性與有效性。接下來，研究將此方法應用在兩個人口流動資料。第一個是從台灣健保資料庫觀察出的民眾就醫地變化。健保資料庫包含了總人口的5%抽樣資料。由於在抽樣上瑕疵不大，因此健保抽樣資料具有了一定的代表性。第二個資料則是英國統計局所提供的人口遷移普查資料。在這兩個資料上，我們發現本研究所提修正法，相較於傳統的空間互動模型具有更好的模型配適表現。此改善程度在非都市地區尤其更為明顯。 / Population movement encompasses various forms, such as migration, mobility, and commuting. Spatial Interaction Model (SIM) serves as an important tool to calibrate these movements in the sense of macro modelling. One of the important features of this model is that the number of migrants often decays with the distance. However, we found that this is not always the case in practice and the decay pattern may change with distance. In this study, we propose a distanced-based modification to the SIM, via applying the techniques of change-point problem to construct distance functional form. Computer simulation is illustrated to validate the method and the empirical analysis of flow data from Taiwan’s National Health Insurance Research Database (NHIRD), and also England & Wales internal migration data also provides sound evidences to support the proposed approach. Note that the flow data from the NHIRD consists of a sample of about one million people and can be treated as a fine sample representative of Taiwan’s whole population (about 23 million people). Our results show that the modified approach is more adequate than the traditional SIM, especially for describing the movements of suburban areas in Taiwan.

L'objectif de ce travail est de comprendre la structure et les interactions des liquides ioniques au contact de surfaces métalliques à l’échelle moléculaire en ayant recours aux méthodes de dynamique moléculaire. Il s’agit également d’étudier l’impact de ces caractéristiques microscopiques sur les propriétés tribologiques du système. Les liquides ioniques choisis en tant qu‘huiles lubrifiantes potentielles présentent des propriétés biodégradables et des caractéristiques tribologiques appropriées. Ils reposent sur des cations alkylammonium combinés avec des anions alkylsulfonate et bistriflamide. Notre étude est structurée en quatre parties. Elle commence par l’analyse des liquides ioniques purs puis, des liquides ioniques confinés entre deux surfaces de fer à l’équilibre et sous cisaillement, et enfin, en présence d’eau. Les propriétés structurales et dynamiques des liquides ioniques sont étudiées à travers la fonction de distribution radiale et les coefficients d’auto-diffusion. L’organisation des charges ainsi que la formation de micro-domaines en solution sont étudiées conjointement au comportement diffusif des espèces ioniques. Un champ de forces atomique, basé sur des méthodes quantiques, a été développé pour modéliser les interactions entre les liquides ioniques et la surface métallique. Des calculs DFT ont été réalisés sur des fragments de liquides ioniques en interaction avec un cluster de fer en fonction de la distance et de leur orientation. Une fonction modélisant des interactions site-site a été ajustée aux valeurs d’énergies fragment–cluster calculées par DFT afin d’obtenir les paramètres du champ de forces. Finalement, la polarisation du métal par les ions a été prise en compte en utilisant un modèle de dipôles induits afin de reproduire l’énergie d’interaction entre les charges et la surface conductrice. Avec ce modèle d’interaction, les simulations de dynamique moléculaire ont permis d’étudier la structure de l’interface entre une surface de fer plane et différents liquides ioniques. Cette analyse s’est concentrée sur l’étude du positionnement des différentes espèces au niveau de la surface, sur l’orientation des chaines alkyles et sur les profils de densité de charge. Des simulations de dynamique moléculaire hors-équilibre de liquides ioniques en interaction avec des surfaces de fer ont été réalisées en utilisant le champ de forces développé précédemment. Un protocole de simulation, basé sur une définition locale de la pression, a été développé pour prédire de manière quantitative le coefficient de friction en fonction de la valeur de la charge et du taux de cisaillement. La dépendance de la friction avec la charge, la vitesse de cisaillement, la topologie de la surface et la taille de la chaine alkyle du liquide ionique a été étudiée. La variation des forces de friction s’explique par l’arrangement spécifique des ions et l’orientation des groupements du liquide ionique à proximité de la surface. Finalement, l’effet de la présence d’eau en petite quantité dans une solution de liquide ionique a aussi été étudié à l’équilibre et hors-équilibre. Un potentiel a été construit pour décrire les interactions entre l’eau et une surface de fer en utilisant la même approche que celle décrite précédemment. Des résultats préliminaires concernant la structure de l’interface liquide-métal et la valeur du coefficient de friction ont été présentés et comparés avec ceux obtenus pour les liquides ioniques purs. / The aim of the present work is to understand at the molecular level the structure and interactions of ionic liquids at metallic surfaces, using molecular dynamics simulations, and to investigate the impact that these microscopic features can have in the tribological properties of the system. The chosen ionic liquids as potential lubricant oils present suitable ecotoxic and biodegradable properties and appropriate tribological characteristics. They are based in alkylammonium cations combined with alkylsulfonate and bistriflamide anions. Our study is divided in four parts, starting from the analyses of pure ionic liquids solutions and evolving to systems of ionic liquids confined between surfaces of iron, at the equilibrium, under shear and also in the presence of water. Structural and dynamic properties of ionic liquids are investigated in terms of the site-site radial distribution functions and the self-diffusion coefficients. The presence of charge-ordering and the formation of micro-domains in solution are discussed, as well as the diffusive behavior of the ionic species. An atomistic force field for ionic liquids interacting with a metal surface was built based on quantum methods. Density functional calculations of alkylammonium cations, alkylsulfonate and bistriflamide anions interacting with a cluster of iron atoms are performed, at a series of distances and orientations. A site-site potential function was then adjusted to the DFT interactions energies, to obtain the force field parameters. Finally, the polarization of the metal by the ions was taken into account using induced dipoles to reproduce the interaction energy between charges and a conductor surface. Using this interaction model, molecular dynamics simulations were performed to study the structure of the interfacial layer of several ionic liquids at a flat iron surface, including analyses of the positional and orientational ordering of the ions near the surface, and charge density profiles. Non-equilibrium molecular dynamics simulations of ionic liquids interacting with iron surfaces were carried out using the specific set of interaction parameters developed previously. A procedure was developed for a quantitative prediction of the friction coefficient at different loads and shear rates, based in a definition of pressure measured locally. The dependence of friction on the load, shear velocity, surface topology and length of alkyl side chains in the ionic liquid was investigated. The changes in the frictional forces were explained in terms of the specific arrangements and orientations of groups forming the ionic liquid at the vicinity of the surface. Finally, the effect of the presence of water in a small quantity in an ionic liquid solution is also studied at equilibrium and non-equilibrium. An interaction potential was build that describes the interaction between water and an iron surface, using the same approach described previously. Preliminary results are presented on the structure at the metal–liquid interface and friction coefficient, and compared with the pure ionic liquids.

Cette thèse est consacrée à la modélisation mathématique du flux sanguin dans les artères en présence de la sténose à cause de l'athérosclérose. L'athérosclérose est une maladie vasculaire complexe caractérisée par la formation d'une plaque menant au rétrécissement de l'artère. Elle est responsable des crises cardiaques et des accidents vasculaires cérébraux. Quels que soient les nombreux facteurs de risque identifiés - cholestérol et lipides, pression, régime alimentaire malsain et obésité - seuls des facteurs mécaniques et hémodynamiques peuvent donner une cause précise de cette maladie. Dans la première partie de la thèse, nous introduisons le modèle mathématique tridimensionnel décrivant l'introduction entre le sang et la paroi artérielle. Le modèle consiste à coupler la dynamique du flux sanguin donnée par les équations de Navier-Stokes formulées dans le cadre Arbitrary Lagrangian Eulerian (ALE) avec les équations élastodynamiques décrivant l'élasticité de la paroi artérielle considérée comme un matériau hyperélastique modélisé par la loi de comportement non-linéaire de Saint Venant-Kirchhoff en tant que système d'interaction fluide-structure. Théoriquement, nous prouvons l'existence et l'unicité locale dans le temps de la solution pour ce système lorsque le fluide est supposé être un fluide homogène Newtonien incompressible et que la structure est décrite par la loi de comportement non-linéaire quasi-incompressible de Saint Venant-Kirchhoff. Les résultats sont établis en utilisant l'outil clé; le théorème du point fixe. La deuxième partie est consacrée à l'analyse numérique de ce modèle. Le sang est considéré comme un fluide non-Newtonien dont le comportement et les propriétés rhéologiques sont décrits par le modèle de Carreau, tandis que la paroi artérielle est un matériau homogène incompressible décrit par les équations élastodynamiques quasi-statiques. Les simulations sont effectuées dans l'espace à deux dimensions R^2 à l'aide du logiciel FreeFem ++ en utilisant la méthode des éléments finis. Nous nous concentrons sur l'étude de la viscosité, de la vitesse et des contraintes de cisaillement maximale. En outre, nous visons à localiser les zones de recirculation qui sont formées à la suite de l'existence de la sténose. En se basant sur de ces résultats, nous procédons à la détection de la zone de solidification où le sang passe de l'état liquide à un matériau de type gelée. Ensuite, nous spécifions que le sang solidifié est un matériau élastique linéaire qui obéit à la loi de Hooke et qui subit à une force de surface externe représentant la contrainte exercée par le sang sur la zone de solidification. Les résultats numériques concernant le sang solidifié sont obtenus en résolvant les équations d'élasticité linéaires à l'aide de FreeFem ++. Nous analysons principalement la déformation de cette zone ainsi que les contraintes de cisaillement la paroi. Les résultats obtenus vont nous permettre de proposer une hypothèse pour la formulation d'un modèle de rupture. / This thesis is devoted to the mathematical modeling of the blood flow in stenosed arteries due to atherosclerosis. Atherosclerosis is a complex vascular disease characterized by the build up of a plaque leading to the narrowing of the artery. It is responsible for heart attacks and strokes. Regardless of the many risk factors that have been identified- cholesterol and lipids, pressure, unhealthy diet and obesity- only mechanical and hemodynamic factors can give a precise cause of this disease. In the first part of the thesis, we introduce the three dimensional mathematical model describing the blood-wall setting. The model consists of coupling the dynamics of the blood flow given by the Navier-Stokes equations formulated in the Arbitrary Lagrangian Eulerian (ALE) framework with the elastodynamic equations describing the elasticity of the arterial wall considered as a hyperelastic material modeled by the non-linear Saint Venant-Kirchhoff model as a fluid-structure interaction (FSI) system. Theoretically, we prove local in time existence and uniqueness of solution for this system when the fluid is assumed to be an incompressible Newtonian homogeneous fluid and the structure is described by the quasi-incompressible non-linear Saint Venant-Kirchhoff model. Results are established relying on the key tool; the fixed point theorem. The second part is devoted for the numerical analysis of the FSI model. The blood is considered to be a non-Newtonian fluid whose behavior and rheological properties are described by Carreau model, while the arterial wall is a homogeneous incompressible material described by the quasi-static elastodynamic equations. Simulations are performed in the two dimensional space R^2 using the finite element method (FEM) software FreeFem++. We focus on investigating the pattern of the viscosity, the speed and the maximum shear stress. Further, we aim to locate the recirculation zones which are formed as a consequence of the existence of the stenosis. Based on these results we proceed to detect the solidification zone where the blood transits from liquid state to a jelly-like material. Next, we specify the solidified blood to be a linear elastic material that obeys Hooke's law and which is subjected to an external surface force representing the stress exerted by the blood on the solidification zone. Numerical results concerning the solidified blood are obtained by solving the linear elasticity equations using FreeFem++. Mainly, we analyze the deformation of this zone as well as the wall shear stress. These analyzed results will allow us to give our hypothesis to derive a rupture model.