Dans la littérature, plusieurs travaux ont été développés autour de la modélisation, l’analyse et la mise en place de politiques de maintenance pour les équipements sujets à des défaillances aléatoires. Ces travaux occultent souvent les réalités industrielles par des hypothèses telles que la connaissance a priori des distributions paramétriques des durées de vie et l’homogénéité des conditions d’exploitation des équipements. Ces hypothèses sont restrictives et constituent une source de biais parce qu’elles conditionnent l’analyse statistique des politiques de maintenance. Dans ce présent travail de thèse, de telles hypothèses sont relaxées pour permettre la prise en compte et la mise en valeurs des informations dérivant directement des données de durées vie issues de l’exploitation de l’équipement et ce sans passer par un modèle paramétrique intermédiaire. L’objectif de ce travail de thèse consiste alors en le développement de modèles statistiques et d’outils efficaces pour l’analyse des politiques de maintenance basées sur les données de durées de vie hétérogènes. Nous proposons en effet une démarche complète d’analyse de stratégies de maintenance en partant des données de durées de vie jusqu’à l’obtention des politiques optimales de maintenance en passant par une phase d’estimation des lois de probabilité. Les politiques de maintenance considérées sont appliques à des équipements usagés évoluant dans des environnements d’exploitation distingués par leur niveau de sévérité. Dans ce contexte, un modèle mathématique est proposé permettant d’évaluer et d’analyser théoriquement les coûts unitaires d’une stratégie de maintenance particulière dite de type âge. Cette analyse a permis d’établir les conditions nécessaires et suffisantes garantissant un âge optimal de remplacement préventif de l’équipement. Les coûts unitaires de maintenance sont complètement estimés par la méthode du Noyau de Parzen. Cette méthode d’estimation est non-paramétrique et définie par une fonction noyau et un paramètre de lissage. Il est également montré, dans nos travaux de recherche, que cet estimateur garantit une faible propagation des erreurs induites par le paramètre de lissage. Les résultats obtenus par la méthode du Noyau de Parzen sont proches des valeurs théoriques avec un faible coefficient de variation. Des extensions de la première politique de maintenance sont également proposées et étudiées. Ce travail de thèse s’achève par la proposition d’une approche permettant de mesurer et d’analyser le risque induit par le report d’une maintenance préventive. Ce risque est analysé à travers une fonction risque proposée / In the reliability literature, several researches works have been developed to deal with modeling, analysis and implementation of maintenance policies for equipments subject to random failures. The majority of these works are based on common assumptions among which the distribution function of the equipment lifetimes is assumed to be known. Furthermore, the equipment is assumed to experience only one operating environment. Such assumptions are indeed restrictive and may introduce a bias in the statistical analysis of the distribution function of the equipment lifetimes which in turn impacts optimization of maintenance policies. In the present research work, these two particular assumptions are relaxed. This relaxation allows to take into account of information related to conditions where the equipment is being operating and to focus on the statistical analysis of maintenance policies without using an intermediate parametric lifetimes distribution. The objective of this thesis consists then on the development of efficient statistical models and tools for managing the maintenance of equipments whose lifetimes distribution is unknown and defined through the heterogeneous lifetimes data. Indeed, this thesis proposes a framework for maintenance strategies determination, from lifetimes data acquisition toward the computation of optimal maintenance policies. The maintenance policies considered are assumed to be performed on used equipments. These later are conduct to experience their missions within different environments each of which is characterized by a degree of severity. In this context, a first mathematical model is proposed to evaluate costs induced by maintenance strategies. The analysis of these costs helps to establish the necessary and sufficient conditions to ensure the existence of an optimal age to perform the preventive maintenance. The maintenance costs are fully estimated by using the Kernel method. This estimation method is non-parametric and defined by two parameters, namely the kernel function and the smoothing parameter. The variability of maintenance costs estimator is deeply analyzed according to the smoothing parameter of Kernel method. From these analyses, it is shown that Kernel estimator method ensures a weak propagation of the errors due to the computation of smoothing parameter. In addition, several simulations are made to estimate the optimal replacement age. These simulations figure out that the numerical results from the Kernel method are close to the theoretical values with a weak coefficient of variation. Two probabilistic extensions of the first mathematical model are proposed and theoretically discussed. To deal with the problem of delayed preventive maintenance, an approach is proposed and discussed. The proposed approach allows evaluating the risk that could induce the delay taken to perform a preventive maintenance at the required optimal date. This approach is based on risk analysis conduct on the basis of a proposed risk function

Classification algorithms have been widely used in many application domains. Most of these domains deal with massive collection of data and hence demand classification algorithms that scale well with the size of the data sets involved. A classification algorithm is said to be scalable if there is no significant increase in time and space requirements for the algorithm (without compromising the generalization performance) when dealing with an increase in the training set size. Support Vector Machine (SVM) is one of the most celebrated kernel based classification methods used in Machine Learning. An SVM capable of handling large scale classification problems will definitely be an ideal candidate in many real world applications. The training process involved in SVM classifier is usually formulated as a Quadratic Programing(QP) problem. The existing solution strategies for this problem have an associated time and space complexity that is (at least) quadratic in the number of training points. This makes the SVM training very expensive even on classification problems having a few thousands of training examples. This thesis addresses the scalability of the training algorithms involved in both two class and multiclass Support Vector Machines. Efficient training schemes reducing the space and time requirements of the SVM training process are proposed as possible solutions. The classification schemes discussed in the thesis for handling large scale two class classification problems are a) Two selective sampling based training schemes for scaling Non-linear SVM and b) Clustering based approaches for handling unbalanced data sets with Core Vector Machine. To handle large scale multicalss classification problems, the thesis proposes Multiclass Core Vector Machine (MCVM), a scalable SVM based multiclass classifier. In MVCM, the multiclass SVM problem is shown to be equivalent to a Minimum Enclosing Ball (MEB) problem and is then solved using a fast approximate MEB finding algorithm. Experimental studies were done with several large real world data sets such as IJCNN1 and Acoustic data sets from LIBSVM page, Extended USPS data set from CVM page and network intrusion detection data sets of DARPA, US Defense used in KDD 99 contest. From the empirical results it is observed that the proposed classification schemes achieve good generalization performance at low time and space requirements. Further, the scalability experiments done with large training data sets have demonstrated that the proposed schemes scale well. A novel soft clustering scheme called Rough Support Vector Clustering (RSVC) employing the idea of Soft Minimum Enclosing Ball Problem (SMEB) is another contribution discussed in this thesis. Experiments done with a synthetic data set and the real world data set namely IRIS, have shown that RSVC finds meaningful soft cluster abstractions.

The application of polynomial reproducing methods has been explored in the context of linear and non linear problems. Of specific interest is the application of a recently developed reproducing scheme, referred to as the error reproducing kernel method (ERKM), which uses non-uniform rational B-splines (NURBS) to construct the basis functions, an aspect that potentially helps bring in locall support, convex approximation and variation diminishing properties in the functional approximation. Polynomial reproducing methods have been applied to solve problems coming under the class of a simplified theory called Cosserat theory. Structures such as a rod which have special geometric properties can be modeled with the aid of such simplified theories. It has been observed that the application of mesh-free methods to solve the aforementioned problems has the advantage that large deformations and exact cross-sectional deformations in a rod could be captured exactly by modeling the rod just in one dimension without the problem of distortion of elements or element locking which would have had some effect if the problem were to be solved using mesh based methods. Polynomial reproducing methods have been applied to problems in fracture mechanics to study the propagation of crack in a structure. As it is often desirable to limit the use of the polynomial reproducing methods to some parts of the domain where their unique advantages such as fast convergence, good accuracy, smooth derivatives, and trivial adaptivity are beneficial, a coupling procedure has been adopted with the objective of using the advantages of both FEM and polynomial reproducing methods. Exploration of SMW (Sherman-Morrison-Woodbury) in the context of polynomial reproducing methods has been done which would assist in calculating the inverse of a perturbed matrix (stiffness matrix in our case). This would to a great extent reduce the cost of computation. In this thesis, as a first step attempts have been made to apply Mesh free cosserat theory to one dimensional problems. The idea was to bring out the advantages and limitations of mesh free cosserat theory and then extend it to 2D problems.

The focus of this thesis is to develop computational techniques for analysis of protein structures. We model protein structures as points in 3-dimensional space which in turn are modeled as weighted graphs. The problem of protein structure comparison is posed as a weighted graph matching problem and an algorithm motivated from the spectral graph matching techniques is developed. The thesis also proposes novel similarity measures by deriving kernel functions. These kernel functions allow the data to be mapped to a suitably defined Reproducing kernel Hilbert Space(RKHS), paving the way for efficient algorithms for protein structure classification. Protein structure comparison (structure alignment)is a classical method of determining overall similarity between two protein structures. This problem can be posed as the approximate weighted subgraph matching problem, which is a well known NP-Hard problem. Spectral graph matching techniques provide efficient heuristic solution for the weighted graph matching problem using eigenvectors of adjacency matrices of the graphs. We propose a novel and efficient algorithm for protein structure comparison using the notion of neighborhood preserving projections (NPP) motivated from spectral graph matching. Empirically, we demonstrate that comparing the NPPs of two protein structures gives the correct equivalences when the sizes of proteins being compared are roughly similar. Also, the resulting algorithm is 3 -20 times faster than the existing state of the art techniques. This algorithm was used for retrieval of protein structures from standard databases with accuracies comparable to the state of the art. A limitation of the above method is that it gives wrong results when the number of unmatched residues, also called insertions and deletions (indels), are very high. This problem was tackled by matching neighborhoods, rather than entire structures. For each pair of neighborhoods, we grow the neighborhood alignments to get alignments for entire structures. This results in a robust method that has outperformed the existing state of the art methods on standard benchmark datasets. This method was also implemented using MPI on a cluster for database search. Another important problem in computational biology is classification of protein structures into classes exhibiting high structural similarity. Many manual and semi-automatic structural classification databases exist. Kernel methods along with support vector machines (SVM) have proved to be a robust and principled tool for classification. We have proposed novel positive semidefinite kernel functions on protein structures based on spatial neighborhoods. The kernels were derived using a general technique called convolution kernel, and showed to be related to the spectral alignment score in a limiting case. These kernels have outperformed the existing tools when validated on a well known manual classification scheme called SCOP. The kernels were designed keeping the general problem of capturing structural similarity in mind, and have been successfully applied to problems in other domains, e.g. computer vision.

Achieving an understanding of the nature of monogenetic volcanic fields depends on identification of the spatial and temporal patterns of volcanism in these fields, and their relationships to structures mapped in the shallow crust and inferred in the deep crust and mantle through interpretation of geochemical, radiometric and geophysical data. We investigate the spatial and temporal distributions of volcanism in the Abu Monogenetic Volcano Group, Southwest Japan. E-W elongated volcano distribution, which is identified by a nonparametric kernel method, is found to be consistent with the spatial extent of P-wave velocity anomalies in the lower crust and upper mantle, supporting the idea that the spatial density map of volcanic vents reflects the geometry of a mantle diapir. Estimated basalt supply to the lower crust is constant. This observation and the spatial distribution of volcanic vents suggest stability of magma productivity and essentially constant two-dimensional size of the source mantle diapir. We mapped conduits, dike segments, and sills in the San Rafael sub-volcanic field, Utah, where the shallowest part of a Pliocene magmatic system is exceptionally well exposed. The distribution of conduits matches the major features of dike distribution, including development of clusters and distribution of outliers. The comparison of San Rafael conduit distribution and the distributions of volcanoes in several recently active volcanic fields supports the use of statistical models, such as nonparametric kernel methods, in probabilistic hazard assessment for distributed volcanism. We developed a new recurrence rate calculation method that uses a Monte Carlo procedure to better reflect and understand the impact of uncertainties of radiometric age determinations on uncertainty of recurrence rate estimates for volcanic activity in the Abu, Yucca Mountain Region, and Izu-Tobu volcanic fields. Results suggest that the recurrence rates of volcanic fields can change by more than one order of magnitude on time scales of several hundred thousand to several million years. This suggests that magma generation rate beneath volcanic fields may change over these time scales. Also, recurrence rate varies more than one order of magnitude between these volcanic fields, consistent with the idea that distributed volcanism may be influenced by both the rate of magma generation and the potential for dike interaction during ascent.

Stochastic and data-distributed optimization algorithms have received lots of attention from the machine learning community due to the tremendous demand from the large-scale learning and the big-data related optimization. A lot of stochastic and deterministic learning algorithms are proposed recently under various application scenarios. Nevertheless, many of these algorithms are based on heuristics and their optimality in terms of the generalization error is not sufficiently justified. In this talk, I will explain the concept of an optimal learning algorithm, and show that given a time budget and proper hypothesis space, only those achieving the lower bounds of the estimation error and the optimization error are optimal. Guided by this concept, we investigated the stochastic minimization of nonsmooth convex loss functions, a central problem in machine learning. We proposed a novel algorithm named Accelerated Nonsmooth Stochastic Gradient Descent, which exploits the structure of common nonsmooth loss functions to achieve optimal convergence rates for a class of problems including SVMs. It is the first stochastic algorithm that can achieve the optimal O(1/t) rate for minimizing nonsmooth loss functions. The fast rates are confirmed by empirical comparisons with state-of-the-art algorithms including the averaged SGD. The Alternating Direction Method of Multipliers (ADMM) is another flexible method to explore function structures. In the second part we proposed stochastic ADMM that can be applied to a general class of convex and nonsmooth functions, beyond the smooth and separable least squares loss used in lasso. We also demonstrate the rates of convergence for our algorithm under various structural assumptions of the stochastic function: O(1/sqrt{t}) for convex functions and O(log t/t) for strongly convex functions. A novel application named Graph-Guided SVM is proposed to demonstrate the usefulness of our algorithm. We also extend the scalability of stochastic algorithms to nonlinear kernel machines, where the problem is formulated as a constrained dual quadratic optimization. The simplex constraint can be handled by the classic Frank-Wolfe method. The proposed stochastic Frank-Wolfe methods achieve comparable or even better accuracies than state-of-the-art batch and online kernel SVM solvers, and are significantly faster. The last part investigates the problem of data-distributed learning. We formulate it as a consensus-constrained optimization problem and solve it with ADMM. It turns out that the underlying communication topology is a key factor in achieving a balance between a fast learning rate and computation resource consumption. We analyze the linear convergence behavior of consensus ADMM so as to characterize the interplay between the communication topology and the penalty parameters used in ADMM. We observe that given optimal parameters, the complete bipartite and the master-slave graphs exhibit the fastest convergence, followed by bi-regular graphs.

Very little is known about the spatio-temporal ecology of the rusty-spotted genet, Genetta maculata. With this study I aimed to describe the activity patterns, resting site use and spatial ecology of G. maculata in Telperion Nature Reserve. I particularly looked at the activity profile and the activity period. I wanted to determine the spatial distribution of resting sites, the number of sites used per individual as well as the index of resting site reuse. I also calculated the distance between resting sites on consecutive days and tested for differences between sexes and seasons. I determined the size of home ranges, as well as that of core areas and compared space use between sexes and seasons as well as vegetation types. A total of six males and nine females were trapped, radio-collared and tracked during continuous night and daytime sessions between September 2015 and August 2016. Rusty-spotted genets were primarily nocturnal (nocturnality index: 0.84) and therefore made use of the darkness for cover when hunting. Overall, male effective activity duration (586 ± 172 min) was greater than for females (564 ± 175 min) possibly because they search for females to mate with as well as due to their larger body size. Seasonal changes in activity were evident – specifically in winter – and were probably a function of both food availability and temperature. Areas with a denser vegetation structure seemed to be more suitable for rusty-spotted genet resting sites. Neither the number of resting sites nor the reuse rate of these resting sites differed between sexes or seasons. The inter-resting site distance on consecutive days was higher for males (938 ± 848 m) than females (707 ± 661 m). This was possibly caused by males travelling larger distances when searching for females to mate with. The inter-resting site distance was higher during autumn, likely due to the decrease in food availability, which made it necessary for genets to increase their hunting efforts. However, a similar increase in hunting effort was not evident during winter as genets decreased their overall activity, possibly in order to avoid colder temperatures. No sexual or seasonal differences in home range size were found. This was attributed to a well-spread and consistent availability of food sources. Core areas only covered on average 7% of the total individual home range which further supports the hypothesis that food was readily available. Both intra- and intersexual home range overlaps were recorded. This was not unusual for carnivores and due to a combination of reproductive and social actions. Home ranges mainly included bushveld vegetation (78%) rather than grassland as these areas provided better cover and likely more abundant food sources. As this was the first exhaustive study of its kind on this species over a full annual cycle, the information gathered is important for the development of conservation strategies for this species, but also for other Genetta species in the rest of Africa. / College of Agriculture and Environmental Sciences / M. Sc. (Environmental Science)

This thesis is related to the Bit Error Rate (BER) estimation for any digital communication system. In many designs of communication systems, the BER is a Key Performance Indicator (KPI). The popular Monte-Carlo (MC) simulation technique is well suited to any system but at the expense of long time simulations when dealing with very low error rates. In this thesis, we propose to estimate the BER by using the Probability Density Function (PDF) estimation of the soft observations of the received bits. First, we have studied a non-parametric PDF estimation technique named the Kernel method. Simulation results in the context of several digital communication systems are proposed. Compared with the conventional MC method, the proposed Kernel-based estimator provides good precision even for high SNR with very limited number of data samples. Second, the Gaussian Mixture Model (GMM), which is a semi-parametric PDF estimation technique, is used to estimate the BER. Compared with the Kernel-based estimator, the GMM method provides better performance in the sense of minimum variance of the estimator. Finally, we have investigated the blind estimation of the BER, which is the estimation when the sent data are unknown. We denote this case as unsupervised BER estimation. The Stochastic Expectation-Maximization (SEM) algorithm combined with the Kernel or GMM PDF estimation methods has been used to solve this issue. By analyzing the simulation results, we show that the obtained BER estimate can be very close to the real values. This is quite promising since it could enable real-time BER estimation on the receiver side without decreasing the user bit rate with pilot symbols for example.