This research is to improve the efficiency of present prediction factors. It has been tested that many factors have prediction power toward stocks returns. Although the prediction power is not stable, the factors are still valuable. This research analyzes preceding factors by neural network in order to make better use of these factors. Besides, we examine 15 companies respectively and compare the results between neural network and liner regression of those companies. Data are divided into training period and prediction period. We use data of training period to build up our model and test the model by the data from prediction period to verify the prediction powers of the models. The results show neural network has better solution compared to liner regression in both training and prediction period. Neural network is more precision and has less prediction error.

Most research on stock prices is based on the present value model or the more general consumption-based model. When applied to real economic data, both of them are found unable to account for both the stock price level and its volatility. Three essays here attempt to both build a more realistic model, and to check whether there is still room for bubbles in explaining fluctuations in stock prices. In the second chapter, several innovations are simultaneously incorporated into the traditional present value model in order to produce more accurate model-based fundamental prices. These innovations comprise replacing with broad dividends the more narrow traditional dividends that are more commonly used, a nonlinear artificial neural network (ANN) forecasting procedure for these broad dividends instead of the more common linear forecasting models for narrow traditional dividends, and a stochastic discount rate in place of the constant discount rate. Empirical results show that the model described above predicts fundamental prices better, compared with alternative models using linear forecasting process, narrow dividends, or a constant discount factor. Nonetheless, actual prices are still largely detached from fundamental prices. The bubble-like deviations are found to coincide with business cycles. The third chapter examines possible cointegration of stock prices with fundamentals and non-fundamentals. The output gap is introduced to form the non-fundamental part of stock prices. I use a trivariate Vector Autoregression (TVAR) model and a single equation model to run cointegration tests between these three variables. Neither of the cointegration tests shows strong evidence of explosive behavior in the DJIA and S&P 500 data. Then, I applied a sup augmented Dickey-Fuller test to check for the existence of periodically collapsing bubbles in stock prices. Such bubbles are found in S&P data during the late 1990s. Employing econometric tests from the third chapter, I continue in the fourth chapter to examine whether bubbles exist in stock prices of conventional economic sectors on the New York Stock Exchange. The ‘old economy’ as a whole is not found to have bubbles. But, periodically collapsing bubbles are found in Material and Telecommunication Services sectors, and the Real Estate industry group.

<p>Evolutionary dynamics can be notoriously complex and difficult to analyze. In this dissertation I describe a population genetic regime where the dynamics are simple enough to allow a relatively complete and elegant treatment. Consider a haploid, asexual population, where each possible genotype has been assigned a fitness. When mutations enter a population sufficiently rarely, we can model the evolution of this population as a Markov chain where the population jumps from one genotype to another at the birth of each new mutant destined for fixation. Furthermore, if the mutation rates are assigned in such a manner that the Markov chain is reversible when all genotypes are assigned the same fitness, then it is still reversible when genotypes are assigned differing fitnesses. </p><p>The key insight is that this Markov chain can be analyzed using the spectral theory of finite-state, reversible Markov chains. I describe the spectral decomposition of the transition matrix and use it to build a general framework with which I address a variety of both classical and novel topics. These topics include a method for creating low-dimensional visualizations of fitness landscapes; a measure of how easy it is for the evolutionary process to `find' a specific genotype or phenotype; the index of dispersion of the molecular clock and its generalizations; a definition for the neighborhood of a genotype based on evolutionary dynamics; and the expected fitness and number of substitutions that have occurred given that a population has been evolving on the fitness landscape for a given period of time. I apply these various analyses to both a simple one-codon fitness landscape and to a large neutral network derived from computational RNA secondary structure predictions.</p> / Dissertation

La connaissance des propriétés diélectriques des tissus biologiques constitue un enjeu majeur pour la santé. Ces propriétés traduisent la manière dont un tissu stocke ou dissipe l’énergie électromagnétique transmise par un champ extérieur ; étant liées à la composition et à la structure du milieu organique, elles traduisent également la nature et l’état physiologique d’un tissu. Leur estimation fine permet donc, le cas échéant, de détecter et/ou de suivre l’évolution d’une pathologie. Parmi les méthodes de caractérisation diélectrique des tissus possibles, nous nous sommes concentrés sur une technique de caractérisation électromagnétique par antenne inductive exploitée en émission/réception, qui permet une mise en œuvre sans contact entre le système de mesure et le tissu. Celle-ci opère dans la gamme des radiofréquences (RF) ce qui présente l’avantage de rendre le dispositif sensible à la fois à la conductivité électrique et à la permittivité diélectrique du tissu. Cette technique travaillant en champ proche nécessite l’utilisation d’un modèle électromagnétique 3D des interactions sonde / tissu pour être mise en œuvre de manière pertinente. Dans ces travaux, nous nous sommes donc intéressés au problème de la modélisation des interactions, ainsi qu’à la résolution du problème inverse qui consiste à estimer les paramètres diélectriques recherchés à partir des données de mesure fournies par l’antenne et du modèle élaboré. Pour cela, nous nous sommes concentrés sur une configuration canonique, constituée d’une antenne RF filiforme circulaire, interagissant avec un milieu diélectrique homogène « sain » dont les paramètres diélectriques macroscopiques sont représentatifs d’un tissu organique (conductivité de 0.6 S/m et permittivité relative de 80), et d’une inclusion sphérique représentative d’une lésion présentant un contraste de 10% à 50% avec les paramètres du milieu « sain ». Nous avons établi un modèle d’interactions électromagnétiques 3D reposant sur une formulation semi – analytique à sources distribuées (DPSM) adaptée à cette configuration. Une étude paramétrique de la mise en œuvre du modèle, validée dans des configurations simples par rapport à des modèles analytiques et des expérimentations, a permis de construire un modèle qui montre des écarts inférieurs à 5 % par rapport à l’expérimentation, et qui établit un compromis acceptable entre exactitude et ressources informatiques nécessaires pour calculer la solution. Enfin, nous nous sommes intéressés à la résolution du problème inverse, consistant à retrouver les paramètres géométriques et diélectriques d’une lésion enfouie dans un milieu diélectrique « sain », à partir des variations d’impédances de l’antenne RF. Pour cela, nous avons construit un modèle inverse à réseaux de neurones artificiels (RNA) à partir de banque de données produites par le modèle DPSM. Une étude paramétrique a permis d’identifier les configurations de mise œuvre (fréquences, positions des antennes) les plus pertinentes permettant d’estimer les propriétés diélectrique, la taille et la position de l’inclusion dans le tissu, avec des erreurs d’estimation de l’ordre de 7% avec une antenne unique monofréquence, pour la caractérisation d’une inclusion de 3 cm de rayon enfouie jusqu’à 6 cm de profondeur. Ces travaux ouvrent la voix à des techniques de diagnostics de dans des milieux plus complexes (tissus stratifiés…) avec des techniques d’investigation multi-antennes et/ou multifréquences particulièrement prometteuses. / The characterization of the dielectric properties of organic tissues is a major issue in health diagnosis. These properties reflect the way organic material stores or dissipates the electromagnetic energy transmitted by an external field. They are related to the composition and the structure of the organic medium. Furthermore, they are also related to the nature and the physiological state of a tissue. For that reason the estimation of these properties is very valuable for detecting and/or monitoring the evolution of tissue pathology.Among the existing dielectric characterization methods, we focused on a characterization technique using an inductive antenna, which acts as a transmitter/receiver sensor and allows a contactless implementation between the measuring system and investigated tissue to be carried out. This system is operated in the radio-frequency (RF) band. Indeed, in the RF the device is equally sensitive to both the electrical conductivity and the dielectric permittivity of the tissue. This technique operates in a near-field and therefore a 3D electromagnetic modeling technique is required to accurately model the interactions between the sensor and the investigated tissue.This work deals with the 3D modeling and with the resolution of the inverse problem required to estimate the dielectric parameters of tissues starting from the data provided by the antenna and the outputs of the model. For this purpose, a canonical configuration featuring a filiform circular antenna is considered. This antenna interacts with a “healthy” homogeneous dielectric medium, which possess the macroscopic dielectric parameters of a typical organic tissue (i.e. conductivity 0.6S/m and relative permittivity of 80 at 100 MHz). Meanwhile, a spherical inclusion buried within the tissue is considered to simulate a tissue lesion. This inclusion features a dielectric contrast of 10% up to 50% by reference to the parameters of the “healthy” medium. A 3D modeling of the sensor/tissue interactions is established, which is based on the distributed point source method (DPSM), a versatile semi-analytical modeling technique. The model is adjusted using a parametric study and validated against analytical models (in simplified configurations) and experiments. The implemented DPSM modeling was found to feature a 5% accuracy error, compared to the experimentations, together with offering an acceptable trade-off between accuracy and the computation cost. Finally, we focused on the solving of the inverse problem which consists in estimating the geometric and dielectric parameters of a buried lesion in the “healthy” dielectric medium, starting from the variations of the impedance of the RF antenna. To do so, a behavioral model build up using an artificial neural network (ANN) was established. The model is build using a data base elaborated using the DPSM model. The parameters of the ANN is discussed in order to identify the relevant configuration (frequency, position of the antenna) to estimate the dielectric properties, the size and the position of the inclusion in the tissue. For a single antenna operated at a single frequency, an inclusion of 3cm radius buried as deep at 6 cm within the tissue was located and characterized with estimation errors of the order of 7%.The methodologies developed in these works open the way to the diagnosis of more complex material (such as layered tissues), using promising techniques such as multi-frequency non contact RF antenna arrays.

Central notions in evolutionary biology are intrinsically topological. This claim is maybe most obvious for the discontinuities associated with punctuated equilibria. Recently, a mathematical framework has been developed that derives the concepts of phenotypic characters and homology from the topological structure of the phenotype space. This structure in turn is determined by the genetic operators and their interplay with the properties of the genotype-phenotype map.

Notre recherche porte sur la recommandation de nouvelles offres d'emploi venant d'être postées et n'ayant pas d'historique d'interactions (démarrage à froid). Nous adaptons les systèmes de recommandations bien connus dans le domaine du commerce électronique à cet objectif, en exploitant les traces d'usage de l'ensemble des demandeurs d'emploi sur les offres antérieures. Une des spécificités du travail présenté est d'avoir considéré des données réelles, et de s'être attaqué aux défis de l'hétérogénéité et du bruit des documents textuels. La contribution présentée intègre l'information des données collaboratives pour apprendre une nouvelle représentation des documents textes, requise pour effectuer la recommandation dite à froid d'une offre nouvelle. Cette représentation dite latente vise essentiellement à construire une bonne métrique. L'espace de recherche considéré est celui des réseaux neuronaux. Les réseaux neuronaux sont entraînés en définissant deux fonctions de perte. La première cherche à préserver la structure locale des informations collaboratives, en s'inspirant des approches de réduction de dimension non linéaires. La seconde s'inspire des réseaux siamois pour reproduire les similarités issues de la matrice collaborative. Le passage à l'échelle de l'approche et ses performances reposent sur l'échantillonnage des paires d'offres considérées comme similaires. L'intérêt de l'approche proposée est démontrée empiriquement sur les données réelles et propriétaires ainsi que sur le benchmark publique CiteULike. Enfin, l'intérêt de la démarche suivie est attesté par notre participation dans un bon rang au challenge international RecSys 2017 (15/100; un million d'utilisateurs pour un million d'offres). / Our research focuses on the recommendation of new job offers that have just been posted and have no interaction history (cold start). To this objective, we adapt well-knowns recommendations systems in the field of e-commerce by exploiting the record of use of all job seekers on previous offers. One of the specificities of the work presented is to have considered real data, and to have tackled the challenges of heterogeneity and noise of textual documents. The presented contribution integrates the information of the collaborative data to learn a new representation of text documents, which is required to make the so-called cold start recommendation of a new offer. The new representation essentially aims to build a good metric. The search space considered is that of neural networks. Neural networks are trained by defining two loss functions. The first seeks to preserve the local structure of collaborative information, drawing on non-linear dimension reduction approaches. The second is inspired by Siamese networks to reproduce the similarities from the collaborative matrix. The scaling up of the approach and its performance are based on the sampling of pairs of offers considered similar. The interest of the proposed approach is demonstrated empirically on the real and proprietary data as well as on the CiteULike public benchmark. Finally, the interest of the approach followed is attested by our participation in a good rank in the international challenge RecSys 2017 (15/100, with millions of users and millions of offers).