Feature extraction is an extremely important pre-processing step to pattern recognition, and machine learning problems. This thesis highlights how one can best extract features from the data in an exhaustively online and purely adaptive manner. The solution to this problem is given for both labeled and unlabeled datasets, by presenting a number of novel on-line learning approaches. Specifically, the differential equation method for solving the generalized eigenvalue problem is used to derive a number of novel machine learning and feature extraction algorithms. The incremental eigen-solution method is used to derive a novel incremental extension of linear discriminant analysis (LDA). Further the proposed incremental version is combined with extreme learning machine (ELM) in which the ELM is used as a preprocessor before learning. In this first key contribution, the dynamic random expansion characteristic of ELM is combined with the proposed incremental LDA technique, and shown to offer a significant improvement in maximizing the discrimination between points in two different classes, while minimizing the distance within each class, in comparison with other standard state-of-the-art incremental and batch techniques. In the second contribution, the differential equation method for solving the generalized eigenvalue problem is used to derive a novel state-of-the-art purely incremental version of slow feature analysis (SLA) algorithm, termed the generalized eigenvalue based slow feature analysis (GENEIGSFA) technique. Further the time series expansion of echo state network (ESN) and radial basis functions (EBF) are used as a pre-processor before learning. In addition, the higher order derivatives are used as a smoothing constraint in the output signal. Finally, an online extension of the generalized eigenvalue problem, derived from James Stone’s criterion, is tested, evaluated and compared with the standard batch version of the slow feature analysis technique, to demonstrate its comparative effectiveness. In the third contribution, light-weight extensions of the statistical technique known as canonical correlation analysis (CCA) for both twinned and multiple data streams, are derived by using the same existing method of solving the generalized eigenvalue problem. Further the proposed method is enhanced by maximizing the covariance between data streams while simultaneously maximizing the rate of change of variances within each data stream. A recurrent set of connections used by ESN are used as a pre-processor between the inputs and the canonical projections in order to capture shared temporal information in two or more data streams. A solution to the problem of identifying a low dimensional manifold on a high dimensional dataspace is then presented in an incremental and adaptive manner. Finally, an online locally optimized extension of Laplacian Eigenmaps is derived termed the generalized incremental laplacian eigenmaps technique (GENILE). Apart from exploiting the benefit of the incremental nature of the proposed manifold based dimensionality reduction technique, most of the time the projections produced by this method are shown to produce a better classification accuracy in comparison with standard batch versions of these techniques - on both artificial and real datasets.

[pt] No presente trabalho propomos uma nova forma de extrair a silhueta de uma nuvem de pontos, via reconstrução local de uma superfície descrita implicitamente por uma função polinomial. Esta reconstrução é baseada nos métodos Gradient one fitting e Ridge regression. A curva silhueta fica definida implicitamente por um sistema de equações não-lineares e sua geração é feita por continuação numérica. Como resultado, verificamos que nosso método se mostrou adequado para tratar dados com ruídos. Além disso, apresentamos um método para a extração local de linhas laplacianas de uma nuvem de pontos baseado na reconstrução local utilizando a triangulação de Delaunay. / [en] In this work we propose a new method for silhouette extraction of a point cloud, via local reconstruction of a surface described implicitly by a polynomial function. This reconstruction is based on the Gradient one fitting and Ridge regression methods. The curve silhouette is implicitly defined by a system of nonlinear equations, and is obtained using numerical continuation. As a result, we observe that our method is suitable to handle noisy data. In addition, we present a method for extracting Laplacian Lines of a point cloud based on local reconstruction using the Delaunay triangulation.

Le travail de ma thèse est basé sur ces 4 points :i) Transport laplacien d'une cellule absorbante :Soit un certain espèce (cellule) de concentration C(x), qui diffuse dans un milieu homogène et isotrope à partir d'une lointaine source localisée sur la frontière fermée $partial Omega_{0}$ vers une interface compact semi-perméable $partial Omega$ (membrane de la "cellule") à laquelle elle disparaisse àun taux d'absorption donné : W>=0. La concentration C (transport laplacien avec un coefficient de diffusion D) satisfaite le problème (P1) (voir la thèse). On s'intéresse à résoudre le problème (P1) en dimension dim = 2; 3 et à calculer les courants local et total à travers les frontières des $partial Omega$ et $partial Omega_{0}$ qui seront utiles pour résoudre le problèmeinverse de localisation. Pour faciliter les calculs et les rendre explicites, on prend $partial Omega$ et $partial Omega_{0}$ avec des formes géométriquement régulières, précisément des boules, en distinguant les deux cas : $Omega$ et $Omega_{0}$ sont concentriques ou non-concentriques. Pour le cas non-concentriques , on utilise la technique de transformation conforme et le développement orthogonal en série de Fourier pour résoudre le problème (P1) en cas bidimensionnel. Tandis que en cas tridimensionnel, on résout le problème (P1) en utilisant le développement orthogonal suivant les fonctions sphériques harmoniques.ii) Problème inverse de localisationOn s'intéresse dans cette partie à résoudre le problème inverse de localisation associé au problème (P1) où les domaines $Omega$ et $Omega_{0}$ sont considérés avec des formes géométriques régulières (précisément des boules) . Ce problème consiste à trouver les conditions de Dirichlet-Neumann sur $partial Omega_{0}$ (courant local, courant total) suffisantes pour déterminer la position de la cellule $partial$ (par rapport à $Omega_{0}$), dont ces conditions sont disponibles par une suite des mesures expérimentales.iii) Problème invesre géomètrique :Dans cette partie on traite un autre type de problème inverse qui consiste à trouver la forme géométrique de la cellule en sachant les conditions de Dirichlet-Neumann au bord extérieur(partial Omega_{0}) qui sont mésurables par une suite d'expérience. Ce type du problème, on l'appelle le problème inverse géométrique. On résout ce problème en utilisant des techniques concernant les fonctions harmoniques et les transformations conformes.iv) Opérateur de Dirichlet-NeumannOn étudie l'opérateur de Dirichlet-Neumann relatif au problème (P1) dans les dimension deux et trois en distinguant les deux cas concentriques et non-concentriques. Ensuite, on montre que cet opérateur de Dirichlet-Neumann engendre certain semi-groupe qu'on l'appelle semi-groupe de Lax. Enfin, on construit ce semi-groupe de Lax associé à cet opérateur en cas tridimensionnel concentriques afin de vérifier que ce semi-groupe admet les mêmes propriétés que celui dans le cas général. / The outline of my thesisi) Let some "species" of concentration C(p), x 2 Rd, diuse stationary in the isotropic bulk from a (distant) source localised on the closed boundary $partial Omega_{0}$ towards a semipermeable compact interface $partial Omega$ of the cell $Omega in Omega_{0}$ where they disappear at a given rate $W >= 0$. Then the steady field of concentrations C satisfy the problem $(P1)$. (see the Thesis). We interest to solve (P1) in Twodimensional and Tridimensional cases and to calculate the local and total flux in order to solving the localisation inverse problem. In order to make easy the calculations, we take $Omega$ and $Omega_{0}$ with a regularly geometricals forms by distinguishing the two cases : Concentrics and non-concentrics case. For the non-cncentrics case, we use the conformal mapping technique for resolving the problem (P1) in the twodimensional case. whereas in the tridimensional case, we use the development according to the spherical harmonics functions.ii) Localisation inverse problemThe aim of the localisation inverse problem is to find the necessary Dirichlet-to-Neumann conditions in order to determine the position of thecell $Omega$, where these conditions are measurable.iii) Geometrical inverse problemOur main results concerns a formal solution of the geometrical inverse problem for the form of absorbing domains. We restrict this study to two dimensions and we study it by the conformal mapping technique and harmonic functions.iv) Dirichlet-to-Neumann operatorWe study the Dirichlet-to-Neumann operatot relative to problem (P1) in the twodimensional and tridimensionnal cases by distinguishing the two cases : Concentrics and non-concentrics case. We prove that the Dirichlet-to-Neumann operator generates some semi-group, we call it the Lax semi-group. Finally we construct this semi group and verify that this demi-group satisfies the generals properties of a operator.

L'objet de cette thèse est de démontrer une formule reliant les métriques de Ray-Singer hypoelliptique et de Milnor sur le déterminant de la cohomologie d'une variété riemannienne compacte par une déformation à la Witten du laplacien hypoelliptique en théorie de de Rham. / The purpose of this thesis is to prove a formula relating the hypoelliptic Ray-Singermetric and the Milnor metric on the determinant of the cohomology of a compact Riemannian manifold by a Witten-like deformation of the hypoelliptic Laplacian in de Rham theory.

Neste trabalho estudamos questões relativas a existência, unicidade, dependência contínua, continuidade, taxas de decaimento e comportamento assintótico de soluções para uma classe de equações de placas lineares e não lineares. No primeiro capítulo revisamos alguns conteúdos e colecionamos uma série de resultados provenientes da teoria geral de análise funcional, semigrupos lineares e atratores, os quais serão aplicados ao longo desta tese. Nos dois próximos capítulos abordamos uma equação da placa de quarta ordem dissipativa com perturbações não lineares do tipo p- Laplaciano e localmente Lipschitz e com memória. No segundo capítulo provamos a estabilidade exponencial de energia correspondente ao problema homogêneo com memória de segunda ordem. Em seguida, no terceiro capítulo estabelecemos resultados que comprovam a existência de um atrator global com dimensão fractal finita para o sistema dinâmico associado ao problema com história de deslocamento relativo que equivale ao problema original. Finalmente, no quarto capítulo tratamos um modelo viscoelástico de placas de Mindlin-Timoshenko de segunda ordem. Nesta ocasião, consideramos essecialmente dois casos, o primeiro quando o sistema é totalmente dissipativo e, em seguida, quando o sistema é parcialmente dissipativo. No primeiro caso, determinamos que o semigrupo linear associado ao problema é analítico e, como consequência, é exponencialmente estável. No segundo caso, mostramos que o semigrupo perde decaimento exponencial e analiticidade, no entanto, provamos que as soluções possuem decaimento do tipo polinomial / In this work we study some questions concerning with existence, uniqueness, continuous dependence, continuity, rates of decay and asymptotic behavior of solutions for a class of linear and nonlinear plate equations. In the first chapter we review some concepts and collect a series of results provided from general theory of functional analysis, linear semigroups and attractors which will be applied throughout this thesis. In the next two chapters we discuss a damped plate equation of fourth order with nonlinear perturbations of the lower order of p-Laplacian type and locally Lipschitz, and a memory term. In the second chapter we prove the exponential stability of energy corresponding to the homogeneous problem with memory of second order. Then in the third chapter we establish some results that allow us to prove the existence of a global attractor with finite fractal dimension for dynamical system associated to the problem with relative displacement history which is equivalent to the original problem. Finally, in the fourth chapter we deal with a viscoelastic Mindlin-Timoshenko plate model of second order. At this moment we consider essentially two cases. The first one when the system is fully damped, then when the system is partially damped. In the first case we show that the semigroup associated to the Mindlin-Timoskenko system is analytic, which in particular implies exponential decay. In the second case we show that such semigroup loses exponential decay, also loses analyticity. However, we prove in this last case that the solutions have decay of polynomial type

Nesse trabalho, usamos métodos da teoria de homogeneização para analisar o compor- tamento assintótico das soluções da equação do p-Laplaciano com condição de contorno de Neumann posto numa família de domínios finos do tipo. De maneira geral, trabalhamos com funções G:(0,1)\\ x R - R uniformemente limitadas, suaves e L-periódicas na segunda variável. Note que o efeito de domínio fino é estabelecido passando ao limite no parâmetro \\varepsilon>0 com \\varepsilon\\to 0. Além disso, introduzimos um parâmetro \\alpha>0 com o objetivo de representar rugosidades via comportamento oscilat\\\'orio na fronteira superior de R^\\varepsilon. Em nossos resultados mostramos que no limite, uma equação unidimensional é obtida, preservando a quasilinearidade do problema original e capturando tanto o efeito da compressão como das oscilações. / In this work we apply homogenization theory methods in order to analyze the asymptotic behavior of the solutions of a p-Laplacian equation with Neumann boundary condition set in bounded thin domains of the type. Generally, we with functions G:(0,1) x R - R uniformly bounded, smooth and L-periodic in the second variable. The thin domain situation is established passing to the limit in the positive parameter \\varepsilon with \\varepsilon \\to 0. In our results we obtain a one dimensional equation that preserves the quasilinearity from the original problem and capturing the effects of compression and oscillations.

The Faber-Krahn theorem states that among all bounded domains with the same volume in Rn (with the standard Euclidean metric), a ball that has lowest first Dirichlet eigenvalue. Recently it has been shown that a similar result holds for (semi-)regular trees. In this article we show that such a theorem also hold for other classes of (not necessarily non-regular) trees. However, for these new results no couterparts in the world of the Laplace-Beltrami-operator on manifolds are known. / Series: Preprint Series / Department of Applied Statistics and Data Processing

Dans cette thèse les théories spectrale et de la diffusion sur des graphes périodiques sont investigué. Le chapitre 1 présente des résultats de préservation de la nature fine du spectre pour des opérateurs de Schrödinger perturbés dans le cadre de cristaux topologiques perturbés. Le chapitre 2 étend ses résultats à des opérateurs du première ordre connu sous le nom de opérateurs de Gauss-Bonnet discrets. Finalement, le chapitre 3 présente des résultats de continuité de composantes spectrales pour des familles de opérateurs de Schrödinger magnétiques sur Z^d / In this thesis we investigate the spectral and scattering theories for crystal lattices. In chapter one we present results concerning the preservation of the nature of the spectrum for perturbed Schrödinger operators acting con perturbed topological crystals. In Chapter 2 we extend this results to some first order operators knowns as discrete Gauss-Bonnet operators. Finally, in chapter 3 we give some results dealing with the continuity of the spectrum for a family of magnetic Schrödinger operators acting on Z^d

Submitted by Rosivalda Pereira (mrs.pereira@ufma.br) on 2017-07-05T21:25:08Z No. of bitstreams: 1 AlanPaciencia.pdf: 382837 bytes, checksum: 5f9c9a1520895e9d9b37a6549ee31251 (MD5) / Made available in DSpace on 2017-07-05T21:25:08Z (GMT). No. of bitstreams: 1 AlanPaciencia.pdf: 382837 bytes, checksum: 5f9c9a1520895e9d9b37a6549ee31251 (MD5) Previous issue date: 2017-01-05 / In this work, we study situations involving the existence, uniqueness, decay rates and asymptotic behavior of solutions for a class of nonlinear equations cards and memory. In particular, in the first chapter we review some issues related to a number of results derived from the general theory of functional analysis, which will be applied during this dissertation. The next chapter will discuss an equation of the fourth order dissipative plate with nonlinear perturbations of type p - Laplacian and locally Lipschitz and memory. Continuing, we prove the exponential stability of energy corresponding to the homogeneous problem with second-order term of memory. / No presente trabalho, estudaremos situações relacionadas a existência, unicidade, taxas de decaimento e comportamentos assintóticos de soluções para uma classe de equações de placas não linear e com termo de memória. Em particular, no primeiro capítulo revisamos alguns assuntos relacionados a uma série de resultados oriundos da teoria geral da análise funcional, os quais ser˜ao aplicados no decorrer dessa dissertação. No capítulo seguinte, abordaremos uma equação da placa de quarta ordem dissipativa com pertubações não lineares do tipo p - Laplaciano e localmente Lipschitz e com termo memória. Continuando, provamos a estabilidade exponencial de energia correspondente ao problema homogêneo com termo de memória de segunda ordem.

abstract: Social media has become the norm of everyone for communication. The usage of social media has increased exponentially in the last decade. The myriads of Social media services such as Facebook, Twitter, Snapchat, and Instagram etc allow people to connect with their friends, and followers freely. The attackers who try to take advantage of this situation has also increased at an exponential rate. Every social media service has its own recommender systems and user profiling algorithms. These algorithms use users current information to make different recommendations. Often the data that is formed from social media services is Linked data as each item/user is usually linked with other users/items. Recommender systems due to their ubiquitous and prominent nature are prone to several forms of attacks. One of the major form of attacks is poisoning the training set data. As recommender systems use current user/item information as the training set to make recommendations, the attacker tries to modify the training set in such a way that the recommender system would benefit the attacker or give incorrect recommendations and hence failing in its basic functionality. Most existing training set attack algorithms work with ``flat" attribute-value data which is typically assumed to be independent and identically distributed (i.i.d.). However, the i.i.d. assumption does not hold for social media data since it is inherently linked as described above. Usage of user-similarity with Graph Regularizer in morphing the training data produces best results to attacker. This thesis proves the same by demonstrating with experiments on Collaborative Filtering with multiple datasets. / Dissertation/Thesis / Masters Thesis Computer Science 2019