Дисертації з теми "Prédiction des séries temporelles multivariées"
Оформте джерело за APA, MLA, Chicago, Harvard та іншими стилями
Ознайомтеся з топ-34 дисертацій для дослідження на тему "Prédiction des séries temporelles multivariées".
Біля кожної праці в переліку літератури доступна кнопка «Додати до бібліографії». Скористайтеся нею – і ми автоматично оформимо бібліографічне посилання на обрану працю в потрібному вам стилі цитування: APA, MLA, «Гарвард», «Чикаго», «Ванкувер» тощо.
Також ви можете завантажити повний текст наукової публікації у форматі «.pdf» та прочитати онлайн анотацію до роботи, якщо відповідні параметри наявні в метаданих.
Переглядайте дисертації для різних дисциплін та оформлюйте правильно вашу бібліографію.
Hmamouche, Youssef. "Prédiction des séries temporelles larges." Thesis, Aix-Marseille, 2018. http://www.theses.fr/2018AIXM0480.
Nowadays, storage and data processing systems are supposed to store and process large time series. As the number of variables observed increases very rapidly, their prediction becomes more and more complicated, and the use of all the variables poses problems for classical prediction models.Univariate prediction models are among the first models of prediction. To improve these models, the use of multiple variables has become common. Thus, multivariate models and become more and more used because they consider more information.With the increase of data related to each other, the application of multivariate models is also questionable. Because the use of all existing information does not necessarily lead to the best predictions. Therefore, the challenge in this situation is to find the most relevant factors among all available data relative to a target variable.In this thesis, we study this problem by presenting a detailed analysis of the proposed approaches in the literature. We address the problem of prediction and size reduction of massive data. We also discuss these approaches in the context of Big Data.The proposed approaches show promising and very competitive results compared to well-known algorithms, and lead to an improvement in the accuracy of the predictions on the data used.Then, we present our contributions, and propose a complete methodology for the prediction of wide time series. We also extend this methodology to big data via distributed computing and parallelism with an implementation of the prediction process proposed in the Hadoop / Spark environment
Labiadh, Mouna. "Méthodologie de construction de modèles adaptatifs pour la simulation énergétique des bâtiments." Thesis, Lyon, 2021. http://www.theses.fr/2021LYSE1158.
Predictive modeling of energy consumption in buildings is essential for intelligent control and efficient planning of energy networks. One way to perform predictive modeling is through machine learning approaches. Alongside their good performance, these approaches are time efficient and facilitates the integration of buildings into smart environments. However, accurate machine learning models rely heavily on collecting relevant building operational data in a sufficient amount, notably when deep learning is used. In the field of buildings energy, historical data are not available for training, such is the case in newly built or newly renovated buildings. Moreover, it is common to verify the energy efficiency of buildings before construction or renovation. For such cases, only a contextual description about the future building and its design is available. The goal of this dissertation is to address the predictive modeling tasks of building energy consumption when no historical data are available for the given target building. To that end, existing data collected from multiple different source buildings are leveraged. This is increasingly relevant with the growth of open data initiatives in various sectors, namely building energy. The main idea is to transfer knowledge across building models. There is little research at the intersection of building energy modeling and knowledge transfer. An important challenge arises when dealing with multi-source data, since large domain shift may exist between different sources and also between each source and the target. As a contribution, a two-fold query-adaptive methodology is developed for cross-building predictive modeling. The first process recommends relevant training data to a target building solely by using a minimal contextual description on it (metadata). Contextual descriptions are provided as user queries. To enable a task-specific recommendation, a deep similarity learning framework is used. The second process trains multiple predictive models based on recommended training data. These models are combined together using an ensemble learning framework to ensure a robust performance. The implementation of the proposed methodology is based on microservices. Logically independent workflows are modeled as microservices with single purposes and separate data sources. Building metadata and time series data collected from multiple sources are integrated into an unified ontology-based view. Experimental evaluation of the predictive model factory validates the effectiveness and the applicability for the use case of building energy modeling. Moreover, because of its generic design, the methodology for query-adaptive cross-domain predictive modeling can be re-used for a diverse range of use cases in different fields
Guerre, Emmanuel. "Méthode non paramétriques d'analyse des séries temporelles multivariées : estimation de mesures de dépendances." Paris 6, 1993. http://www.theses.fr/1993PA066110.
Ziat, Ali Yazid. "Apprentissage de représentation pour la prédiction et la classification de séries temporelles." Thesis, Paris 6, 2017. http://www.theses.fr/2017PA066324/document.
This thesis deals with the development of time series analysis methods. Our contributions focus on two tasks: time series forecasting and classification. Our first contribution presents a method of prediction and completion of multivariate and relational time series. The aim is to be able to simultaneously predict the evolution of a group of time series connected to each other according to a graph, as well as to complete the missing values in these series (which may correspond for example to a failure of a sensor during a given time interval). We propose to use representation learning techniques to forecast the evolution of the series while completing the missing values and taking into account the relationships that may exist between them. Extensions of this model are proposed and described: first in the context of the prediction of heterogeneous time series and then in the case of the prediction of time series with an expressed uncertainty. A prediction model of spatio-temporal series is then proposed, in which the relations between the different series can be expressed more generally, and where these can be learned.Finally, we are interested in the classification of time series. A joint model of metric learning and time-series classification is proposed and an experimental comparison is conducted
Harlé, Flore. "Détection de ruptures multiples dans des séries temporelles multivariées : application à l'inférence de réseaux de dépendance." Thesis, Université Grenoble Alpes (ComUE), 2016. http://www.theses.fr/2016GREAT043/document.
This thesis presents a method for the multiple change-points detection in multivariate time series, and exploits the results to estimate the relationships between the components of the system. The originality of the model, called the Bernoulli Detector, relies on the combination of a local statistics from a robust test, based on the computation of ranks, with a global Bayesian framework. This non parametric model does not require strong hypothesis on the distribution of the observations. It is applicable without modification on gaussian data as well as data corrupted by outliers. The detection of a single change-point is controlled even for small samples. In a multivariate context, a term is introduced to model the dependencies between the changes, assuming that if two components are connected, the events occurring in the first one tend to affect the second one instantaneously. Thanks to this flexible model, the segmentation is sensitive to common changes shared by several signals but also to isolated changes occurring in a single signal. The method is compared with other solutions of the literature, especially on real datasets of electrical household consumption and genomic measurements. These experiments enhance the interest of the model for the detection of change-points in independent, conditionally independent or fully connected signals. The synchronization of the change-points within the time series is finally exploited in order to estimate the relationships between the variables, with the Bayesian network formalism. By adapting the score function of a structure learning method, it is checked that the independency model that describes the system can be partly retrieved through the information given by the change-points, estimated by the Bernoulli Detector
Coelho, rodrigues Pedro Luiz. "Exploration des invariances de séries temporelles multivariées via la géométrie Riemannienne : validation sur des données EEG." Thesis, Université Grenoble Alpes (ComUE), 2019. http://www.theses.fr/2019GREAT095.
Multivariate time series are the standard tool for describing and analysing measurements from multiple sensors during an experiment. In this work, we discuss different aspects of such representations that are invariant to transformations occurring in practical situations. The main source of inspiration for our investigations are experiments with neural signals from electroencephalography (EEG), but the ideas that we present are amenable to other kinds of time series.The first invariance that we consider concerns the dimensionality of the multivariate time series. Very often, signals recorded from neighbouring sensors present strong statistical dependency between them. We present techniques for disposing of the redundancy of these correlated signals and obtaining new multivariate time series that represent the same phenomenon but in a smaller dimension.The second invariance that we treat is related to time series describing the same phenomena but recorded under different experimental conditions. For instance, signals recorded with the same experimental apparatus but on different days of the week, different test subjects, etc. In such cases, despite an underlying variability, the multivariate time series share certain commonalities that can be exploited for joint analysis. Moreover, reusing information already available from other datasets is a very appealing idea and allows for “data-efficient” machine learning methods. We present an original transfer learning procedure that transforms these time series so that their statistical distributions become aligned and can be pooled together for further statistical analysis.Finally, we extend the previous case to when the time series are obtained from different experimental conditions and also different experimental setups. A practical example is having EEG recordings from subjects executing the same cognitive task but with the electrodes positioned differently. We present an original method that transforms these multivariate time series so that they become compatible in terms of dimensionality and also in terms of statistical distributions.We illustrate the techniques described above on EEG epochs recorded during brain-computer interface (BCI) experiments. We show examples where the reduction of the multivariate time series does not affect the performance of statistical classifiers used to distinguish their classes, as well as instances where our transfer learning and dimension-matching proposals provide remarkable results on classification in cross-session and cross-subject settings.For exploring the invariances presented above, we rely on a framework that parametrizes the statistics of the multivariate time series via Hermitian positive definite (HPD) matrices. We manipulate these matrices by considering them in a Riemannian manifold in which an adequate metric is chosen. We use concepts from Riemannian geometry to define notions such as geodesic distance, center of mass, and statistical classifiers for time series. This approach is rooted on fundamental results of differential geometry for Hermitian positive definite matrices and has links with other well established areas in applied mathematics, such as information geometry and signal processing
Plaud, Angéline. "Classification ensembliste des séries temporelles multivariées basée sur les M-histogrammes et une approche multi-vues." Thesis, Université Clermont Auvergne (2017-2020), 2019. http://www.theses.fr/2019CLFAC047.
Recording measurements about various phenomena and exchanging information about it, participate in the emergence of a type of data called time series. Today humongous quantities of those data are often collected. A time series is characterized by numerous points and interactions can be observed between those points. A time series is multivariate when multiple measures are recorded at each timestamp, meaning a point is, in fact, a vector of values. Even if univariate time series, one value at each timestamp, are well-studied and defined, it’s not the case of multivariate one, for which the analysis is still challenging. Indeed, it is not possible to apply directly techniques of classification developed on univariate data to the case of multivariate one. In fact, for this latter, we have to take into consideration the interactions not only between points but also between dimensions. Moreover, in industrial cases, as in Michelin company, the data are big and also of different length in terms of points size composing the series. And this brings a new complexity to deal with during the analysis. None of the current techniques of classifying multivariate time series satisfies the following criteria, which are a low complexity of computation, dealing with variation in the number of points and good classification results. In our approach, we explored a new tool, which has not been applied before for MTS classification, which is called M-histogram. A M-histogram is a visualization tool using M axis to project the density function underlying the data. We have employed it here to produce a new representation of the data, that allows us to bring out the interactions between dimensions. Searching for links between dimensions correspond particularly to a part of learning techniques called multi-view learning. A view is an extraction of dimensions of a dataset, which are of same nature or type. Then the goal is to display the links between the dimensions inside each view in order to classify all the data, using an ensemble classifier. So we propose a multi-view ensemble model to classify multivariate time series. The model creates multiple M-histograms from differents groups of dimensions. Then each view allows us to get a prediction which we can aggregate to get a final prediction. In this thesis, we show that the proposed model allows a fast classification of multivariate time series of different sizes. In particular, we applied it on aMichelin use case
Vroman, Philippe. "Prédiction des séries temporelles en milieu incertain : application à la prévision de ventes dans la distribution textile." Lille 1, 2000. http://www.theses.fr/2000LIL10207.
Arnoux, Thibaud. "Prédiction d'interactions dans les flots de liens. Combiner les caractéristiques structurelles et temporelles." Thesis, Sorbonne université, 2018. http://www.theses.fr/2018SORUS229.
The link stream formalism represent an approach allowing to capture the system dynamic while providing a framework to understand the system's behavior. A link stream is a sequence of triplet (t,u,v) indicating that an interaction occurred between u and v at time t. The importance of the system's dynamic during the prediction places it at the crossroads of link prediction in graphs and time series prediction. We will explore several formalizations of the problem of prediction in link streams. In the following we will study the activity prediction, that is to say predicting the number of interactions occurring in the future between each pair of nodes during a given period. We introduce the protocol, allowing to combine the data characteristics to predict the activity. We study the behavior of our protocol during several experiments on four datasets et evaluate the prediction quality. We will look at how the introduction of pair of nodes classes allows to preserve the link diversity in the prediction while improving the prediction. Our goal is to define a general prediction framework allowing in-depth studies of the relationship between temporal and structural characteristics in prediction tasks
Ahmad, Ali. "Contribution à l'économétrie des séries temporelles à valeurs entières." Thesis, Lille 3, 2016. http://www.theses.fr/2016LIL30059/document.
The framework of this PhD dissertation is the conditional mean count time seriesmodels. We propose the Poisson quasi-maximum likelihood estimator (PQMLE) for the conditional mean parameters. We show that, under quite general regularityconditions, this estimator is consistent and asymptotically normal for a wide classeof count time series models. Since the conditional mean parameters of some modelsare positively constrained, as, for example, in the integer-valued autoregressive (INAR) and in the integer-valued generalized autoregressive conditional heteroscedasticity (INGARCH), we study the asymptotic distribution of this estimator when the parameter lies at the boundary of the parameter space. We deduce a Waldtype test for the significance of the parameters and another Wald-type test for the constance of the conditional mean. Subsequently, we propose a robust and general goodness-of-fit test for the count time series models. We derive the joint distribution of the PQMLE and of the empirical residual autocovariances. Then, we deduce the asymptotic distribution of the estimated residual autocovariances and also of a portmanteau test. Finally, we propose the PQMLE for estimating, equation-by-equation (EbE), the conditional mean parameters of a multivariate time series of counts. By using slightly different assumptions from those given for PQMLE, we show the consistency and the asymptotic normality of this estimator for a considerable variety of multivariate count time series models
Lepère, Stéphane. "Contribution à la prédiction en ligne des séries temporelles : un cas d'étude à la modélisation de systèmes dynamiques." Lille 1, 2001. https://pepite-depot.univ-lille.fr/RESTREINT/Th_Num/2001/50376-2001-219.pdf.
Khodor, Nadine. "Analyse de la dynamique des séries temporelles multi-variées pour la prédiction d’une syncope lors d’un test d’inclinaison." Thesis, Rennes 1, 2014. http://www.theses.fr/2014REN1S123/document.
Syncope is a sudden loss of consciousness. Although it is not usually fatal, it has an economic impact on the health care system and the personal lives of people suffering. The purpose of this study is to reduce the duration of the clinical test (approximately 1 hour) and to avoid patients to develop syncope by early predicting the occurrence of syncope. The entire work fits into a data mining approach involving the feature extraction, feature selection and classification. 3 complementary approaches are proposed, the first one exploits nonlinear analysis methods of time series extracted from signals acquired during the test, the second one focuses on time- frequency (TF) relation between signals and suggests new indexes and the third one, the most original, takes into account their temporal dynamics
Marquez, Alfonzo Bicky. "Reservoir computing photonique et méthodes non-linéaires de représentation de signaux complexes : Application à la prédiction de séries temporelles." Thesis, Bourgogne Franche-Comté, 2018. http://www.theses.fr/2018UBFCD042/document.
Artificial neural networks are systems prominently used in computation and investigations of biological neural systems. They provide state-of-the-art performance in challenging problems like the prediction of chaotic signals. Yet, the understanding of how neural networks actually solve problems like prediction remains vague; the black-box analogy is often employed. Merging nonlinear dynamical systems theory with machine learning, we develop a new concept which describes neural networks and prediction within the same framework. Taking profit of the obtained insight, we a-priori design a hybrid computer, which extends a neural network by an external memory. Furthermore, we identify mechanisms based on spatio-temporal synchronization with which random recurrent neural networks operated beyond their fixed point could reduce the negative impact of regular spontaneous dynamics on their computational performance. Finally, we build a recurrent delay network in an electro-optical setup inspired by the Ikeda system, which at first is investigated in a nonlinear dynamics framework. We then implement a neuromorphic processor dedicated to a prediction task
Voyant, Cyril. "Prédiction de séries temporelles de rayonnement solaire global et de production d'énergie photovoltaïque à partir de réseaux de neurones artificiels." Phd thesis, Université Pascal Paoli, 2011. http://tel.archives-ouvertes.fr/tel-00635298.
Sànchez, Pérez Andrés. "Agrégation de prédicteurs pour des séries temporelles, optimalité dans un contexte localement stationnaire." Thesis, Paris, ENST, 2015. http://www.theses.fr/2015ENST0051/document.
This thesis regroups our results on dependent time series prediction. The work is divided into three main chapters where we tackle different problems. The first one is the aggregation of predictors of Causal Bernoulli Shifts using a Bayesian approach. The second one is the aggregation of predictors of what we define as sub-linear processes. Locally stationary time varying autoregressive processes receive a particular attention; we investigate an adaptive prediction scheme for them. In the last main chapter we study the linear regression problem for a general class of locally stationary processes
Huard, Malo. "Apprentissage et prévision séquentiels : bornes uniformes pour le regret linéaire et séries temporelles hiérarchiques." Thesis, université Paris-Saclay, 2020. http://www.theses.fr/2020UPASM009.
This work presents some theoretical and practical contributions to the prediction of arbitrary sequences. In this domain, forecasting takes place sequentially at the same time as learning. At each step, the model is fitted on the past data in order to predict the next observation. The goal of this model is to make the best possible predictions, i.e. those that minimize their deviations from the observations, which are made a posteriori. Sequential learning methods are evaluated by their regret, which measures how close strategies are to the best possible, known only after all the data is available. In this thesis, we extend the set of weights vectors a method is compared to when doing sequential linear regression. We have adapted an existing algorithm by improving its theoretical guarantees allowing it to be compared to any constant linear combination without restriction on the norm of its mixing weights. A second work consisted in extending sequential forecasting methods when forcasted data is organized in a hierarchy. We tested these hierarchical methods on two practical applications, household power consumption prediction and demand forecasts in e-commerce
Çinar, Yagmur Gizem. "Prédiction de séquences basée sur des réseaux de neurones récurrents dans le contexte des séries temporelles et des sessions de recherche d'information." Thesis, Université Grenoble Alpes (ComUE), 2019. http://www.theses.fr/2019GREAM079.
This thesis investigates challenges of sequence prediction in different scenarios such as sequence prediction using recurrent neural networks (RNNs) in the context of time series and information retrieval (IR) search sessions. Predicting the unknown values that follow some previously observed values is basically called sequence prediction.It is widely applicable to many domains where a sequential behavior is observed in the data. In this study, we focus on two different types of sequence prediction tasks: time series forecasting and next query prediction in an information retrieval search session.Time series often display pseudo-periods, i.e. time intervals with strong correlation between values of time series. Seasonal changes in weather time series or electricity usage at day and night time are some examples of pseudo-periods. In a forecasting scenario, pseudo-periods correspond to the difference between the positions of the output being predicted and specific inputs.In order to capture periods in RNNs, one needs a memory of the input sequence. Sequence-to-sequence RNNs (with attention mechanism) reuse specific (representations of) input values to predict output values. Sequence-to-sequence RNNs with an attention mechanism seem to be adequate for capturing periods. In this manner, we first explore the capability of an attention mechanism in that context. However, according to our initial analysis, a standard attention mechanism did not perform well to capture the periods. Therefore, we propose a period-aware content-based attention RNN model. This model is an extension of state-of-the-art sequence-to-sequence RNNs with attention mechanism and it is aimed to capture the periods in time series with or without missing values.Our experimental results with period-aware content-based attention RNNs show significant improvement on univariate and multivariate time series forecasting performance on several publicly available data sets.Another challenge in sequence prediction is the next query prediction. The next query prediction helps users to disambiguate their search query, to explore different aspects of the information they need or to form a precise and succint query that leads to higher retrieval performance. A search session is dynamic, and the information need of a user might change over a search session as a result of the search interactions. Furthermore, interactions of a user with a search engine influence the user's query reformulations. Considering this influence on the query formulations, we first analyze where the next query words come from? Using the analysis of the sources of query words, we propose two next query prediction approaches: a set view and a sequence view.The set view adapts a bag-of-words approach using a novel feature set defined based on the sources of next query words analysis. Here, the next query is predicted using learning to rank. The sequence view extends a hierarchical RNN model by considering the sources of next query words in the prediction. The sources of next query words are incorporated by using an attention mechanism on the interaction words. We have observed using sequence approach, a natural formulation of the problem, and exploiting all sources of evidence lead to better next query prediction
Germain, Simon. "Conception d'une mesure automatisée de détection des changements alimentaires chez le porc." Mémoire, Université de Sherbrooke, 2015. http://hdl.handle.net/11143/7925.
Rynkiewicz, Joseph. "Modèles hybrides intégrant des réseaux de neurones artificiels à des modèles de chaînes de Markov cachées : application à la prédiction de séries temporelles." Paris 1, 2000. http://www.theses.fr/2000PA010077.
Vuillemin, Benoit. "Recherche de règles de prédiction dans un contexte d'Intelligence Ambiante." Thesis, Lyon, 2020. http://www.theses.fr/2020LYSE1120.
This thesis deals with the subject of Ambient Intelligence, the fusion between Artificial Intelligence and the Internet of Things. The goal of this work is to extract prediction rules from the data provided by connected objects in an environment, in order to propose automation to users. Our main concern relies on privacy, user interactions, and the explainability of the system’s operation. In this context, several contributions were made. The first is an ambient intelligence architecture that operates locally, and processes data from a single connected environment. The second is a discretization process without a priori on the input data, allowing to take into account different kinds of data from various objects. The third is a new algorithm for searching rules over a time series, which avoids the limitations of stateoftheart algorithms. The approach was validated by tests on two real databases. Finally, prospects for future developments in the system are presented
Mousheimish, Raef. "Combinaison de l’Internet des objets, du traitement d’évènements complexes et de la classification de séries temporelles pour une gestion proactive de processus métier." Thesis, Université Paris-Saclay (ComUE), 2017. http://www.theses.fr/2017SACLV073/document.
Internet of things is at the core ofsmart industrial processes thanks to its capacityof event detection from data conveyed bysensors. However, much remains to be done tomake the most out of this recent technologyand make it scale. This thesis aims at filling thegap between the massive data flow collected bysensors and their effective exploitation inbusiness process management. It proposes aglobal approach, which combines stream dataprocessing, supervised learning and/or use ofcomplex event processing rules allowing topredict (and thereby avoid) undesirable events,and finally business process managementextended to these complex rules. The scientificcontributions of this thesis lie in several topics:making the business process more intelligentand more dynamic; automation of complexevent processing by learning the rules; and lastand not least, in datamining for multivariatetime series by early prediction of risks. Thetarget application of this thesis is theinstrumented transportation of artworks
Pineau, Edouard. "Contributions to representation learning of multivariate time series and graphs." Thesis, Institut polytechnique de Paris, 2020. http://www.theses.fr/2020IPPAT037.
Machine learning (ML) algorithms are designed to learn models that have the ability to take decisions or make predictions from data, in a large panel of tasks. In general, the learned models are statistical approximations of the true/optimal unknown decision models. The efficiency of a learning algorithm depends on an equilibrium between model richness, complexity of the data distribution and complexity of the task to solve from data. Nevertheless, for computational convenience, the statistical decision models often adopt simplifying assumptions about the data (e.g. linear separability, independence of the observed variables, etc.). However, when data distribution is complex (e.g. high-dimensional with nonlinear interactions between observed variables), the simplifying assumptions can be counterproductive. In this situation, a solution is to feed the model with an alternative representation of the data. The objective of data representation is to separate the relevant information with respect to the task to solve from the noise, in particular if the relevant information is hidden (latent), in order to help the statistical model. Until recently and the rise of modern ML, many standard representations consisted in an expert-based handcrafted preprocessing of data. Recently, a branch of ML called deep learning (DL) completely shifted the paradigm. DL uses neural networks (NNs), a family of powerful parametric functions, as learning data representation pipelines. These recent advances outperformed most of the handcrafted data in many domains.In this thesis, we are interested in learning representations of multivariate time series (MTS) and graphs. MTS and graphs are particular objects that do not directly match standard requirements of ML algorithms. They can have variable size and non-trivial alignment, such that comparing two MTS or two graphs with standard metrics is generally not relevant. Hence, particular representations are required for their analysis using ML approaches. The contributions of this thesis consist of practical and theoretical results presenting new MTS and graphs representation learning frameworks.Two MTS representation learning frameworks are dedicated to the ageing detection of mechanical systems. First, we propose a model-based MTS representation learning framework called Sequence-to-graph (Seq2Graph). Seq2Graph assumes that the data we observe has been generated by a model whose graphical representation is a causality graph. It then represents, using an appropriate neural network, the sample on this graph. From this representation, when it is appropriate, we can find interesting information about the state of the studied mechanical system. Second, we propose a generic trend detection method called Contrastive Trend Estimation (CTE). CTE learns to classify pairs of samples with respect to the monotony of the trend between them. We show that using this method, under few assumptions, we identify the true state underlying the studied mechanical system, up-to monotone scalar transform.Two graph representation learning frameworks are dedicated to the classification of graphs. First, we propose to see graphs as sequences of nodes and create a framework based on recurrent neural networks to represent and classify them. Second, we analyze a simple baseline feature for graph classification: the Laplacian spectrum. We show that this feature matches minimal requirements to classify graphs when all the meaningful information is contained in the structure of the graphs
Fauvel, Kevin. "Enhancing performance and explainability of multivariate time series machine learning methods : applications for social impact in dairy resource monitoring and earthquake early warning." Thesis, Rennes 1, 2020. http://www.theses.fr/2020REN1S043.
The prevalent deployment and usage of sensors in a wide range of sectors generate an abundance of multivariate data which has proven to be instrumental for researches, businesses and policies. More specifically, multivariate data which integrates temporal evolution, i.e. Multivariate Time Series (MTS), has received significant interests in recent years, driven by high resolution monitoring applications (e.g. healthcare, mobility) and machine learning. However, for many applications, the adoption of machine learning methods cannot rely solely on their prediction performance. For example, the European Union’s General Data Protection Regulation, which became enforceable on 25 May 2018, introduces a right to explanation for all individuals so that they can obtain “meaningful explanations of the logic involved” when automated decision-making has “legal effects” on individuals or similarly “significantly affecting” them. The current best performing state-of-the-art MTS machine learning methods are “black-box” models, i.e. complicated-to-understand models, which rely on explainability methods providing explanations from any machine learning model to support their predictions (post-hoc model-agnostic). The main line of work in post-hoc model-agnostic explainability methods approximates the decision surface of a model using an explainable surrogate model. However, the explanations from the surrogate models cannot be perfectly faithful with respect to the original model, which is a prerequisite for numerous applications. Faithfulness is critical as it corresponds to the level of trust an end-user can have in the explanations of model predictions, i.e. the level of relatedness of the explanations to what the model actually computes. This thesis introduces new approaches to enhance both performance and explainability of MTS machine learning methods, and derive insights from the new methods about two real-world applications
Moudiki, Thierry. "Interest rates modeling for insurance : interpolation, extrapolation, and forecasting." Thesis, Lyon, 2018. http://www.theses.fr/2018LYSE1110/document.
The Own Risk Solvency and Assessment (ORSA) is a set of processes defined by the European prudential directive Solvency II, that serve for decision-making and strategic analysis. In the context of ORSA, insurance companies are required to assess their solvency needs in a continuous and prospective way. For this purpose, they notably need to forecast their balance sheet -asset and liabilities- over a defined horizon. In this work, we specifically focus on the asset forecasting part. This thesis is about the Yield Curve, Forecasting, and Forecasting the Yield Curve. We present a few novel techniques for the construction, the extrapolation of static curves (that is, curves which are constructed at a fixed date), and for forecasting the spot interest rates over time. Throughout the text, when we say "Yield Curve", we actually mean "Discount curve". That is: we ignore the counterparty credit risk, and consider that the curves are risk-free. Though, the same techniques could be applied to construct/forecast the actual risk-free curves and credit spread curves, and combine both to obtain pseudo- discount curves incorporating the counterparty credit risk
Fries, Sébastien. "Anticipative alpha-stable linear processes for time series analysis : conditional dynamics and estimation." Thesis, Université Paris-Saclay (ComUE), 2018. http://www.theses.fr/2018SACLG005/document.
In the framework of linear time series analysis, we study a class of so-called anticipative strictly stationary processes potentially depending on all the terms of an independent and identically distributed alpha-stable errors sequence.Focusing first on autoregressive (AR) processes, it is shown that higher order conditional moments than marginal ones exist provided the characteristic polynomials admits at least one root inside the unit circle. The forms of the first and second order moments are obtained in special cases.The least squares method is shown to provide a consistent estimator of an all-pass causal representation of the process, the validity of which can be tested by a portmanteau-type test. A method based on extreme residuals clustering is proposed to determine the original AR representation.The anticipative stable AR(1) is studied in details in the framework of bivariate alpha-stable random vectors and the functional forms of its first four conditional moments are obtained under any admissible parameterisation.It is shown that during extreme events, these moments become equivalent to those of a two-point distribution charging two polarly-opposite future paths: exponential growth or collapse.Parallel results are obtained for the continuous time counterpart of the AR(1), the anticipative stable Ornstein-Uhlenbeck process.For infinite alpha-stable moving averages, the conditional distribution of future paths given the observed past trajectory during extreme events is derived on the basis of a new representation of stable random vectors on unit cylinders relative to semi-norms.Contrary to the case of norms, such representation yield a multivariate regularly varying tails property appropriate for prediction purposes, but not all stable vectors admit such a representation.A characterisation is provided and it is shown that finite length paths of a stable moving average admit such representation provided the process is "anticipative enough".Processes resulting from the linear combination of stable moving averages are encompassed, and the conditional distribution has a natural interpretation in terms of pattern identification
Phan, Thi-Thu-Hong. "Elastic matching for classification and modelisation of incomplete time series." Thesis, Littoral, 2018. http://www.theses.fr/2018DUNK0483/document.
Missing data are a prevalent problem in many domains of pattern recognition and signal processing. Most of the existing techniques in the literature suffer from one major drawback, which is their inability to process incomplete datasets. Missing data produce a loss of information and thus yield inaccurate data interpretation, biased results or unreliable analysis, especially for large missing sub-sequence(s). So, this thesis focuses on dealing with large consecutive missing values in univariate and low/un-correlated multivariate time series. We begin by investigating an imputation method to overcome these issues in univariate time series. This approach is based on the combination of shape-feature extraction algorithm and Dynamic Time Warping method. A new R-package, namely DTWBI, is then developed. In the following work, the DTWBI approach is extended to complete large successive missing data in low/un-correlated multivariate time series (called DTWUMI) and a DTWUMI R-package is also established. The key of these two proposed methods is that using the elastic matching to retrieving similar values in the series before and/or after the missing values. This optimizes as much as possible the dynamics and shape of knowledge data, and while applying the shape-feature extraction algorithm allows to reduce the computing time. Successively, we introduce a new method for filling large successive missing values in low/un-correlated multivariate time series, namely FSMUMI, which enables to manage a high level of uncertainty. In this way, we propose to use a novel fuzzy grades of basic similarity measures and fuzzy logic rules. Finally, we employ the DTWBI to (i) complete the MAREL Carnot dataset and then we perform a detection of rare/extreme events in this database (ii) forecast various meteorological univariate time series collected in Vietnam
Andreux, Mathieu. "Foveal autoregressive neural time-series modeling." Thesis, Paris Sciences et Lettres (ComUE), 2018. http://www.theses.fr/2018PSLEE073.
This dissertation studies unsupervised time-series modelling. We first focus on the problem of linearly predicting future values of a time-series under the assumption of long-range dependencies, which requires to take into account a large past. We introduce a family of causal and foveal wavelets which project past values on a subspace which is adapted to the problem, thereby reducing the variance of the associated estimators. We then investigate under which conditions non-linear predictors exhibit better performances than linear ones. Time-series which admit a sparse time-frequency representation, such as audio ones, satisfy those requirements, and we propose a prediction algorithm using such a representation. The last problem we tackle is audio time-series synthesis. We propose a new generation method relying on a deep convolutional neural network, with an encoder-decoder architecture, which allows to synthesize new realistic signals. Contrary to state-of-the-art methods, we explicitly use time-frequency properties of sounds to define an encoder with the scattering transform, while the decoder is trained to solve an inverse problem in an adapted metric
Beaufils, Bertrand. "Topological Data Analysis and Statistical Learning for measuring pedestrian activities from inertial sensors." Thesis, université Paris-Saclay, 2020. http://www.theses.fr/2020UPASS107.
This thesis focuses on the detection of specific movements using ActiMyo, a device developed by the company Sysnav. This system is composed by low-cost miniature inertial sensors that can be worn on the ankle and wrist. In particular, a supervised statistical learning approach aims to detect strides in ankle recordings. This first work, combined with an algorithm patented by Sysnav, allows to compute the trajectory of the pedestrian. This trajectory is then used in a new supervised learning method for the activity recognition, which is valuable information, especially in a medical context. These two algorithms offer an innovative approach based on the alignment of inertial signals and the extraction of candidate intervals which are then classified by the Gradient Boosting Trees algorithm. This thesis also presents a neural network architecture combining convolutional channels and topological data analysis for the detection of movements representative of Parkinson’s disease such as tremors and dyskinesia crises
Schettini, Frédéric. "Fusion de données pour la surveillance du trafic et l'information des usagers." Toulouse, ENSAE, 1998. http://www.theses.fr/1998ESAE0016.
De, Montera Louis. "Etude de la variabilité micro-échelle des précipitations : Application à la propagation des ondes millimétriques en SATCOM." Phd thesis, Université de Versailles-Saint Quentin en Yvelines, 2008. http://tel.archives-ouvertes.fr/tel-00384050.
Afin de prédire l'affaiblissement sur la liaison montante à partir de celui sur la liaison descendante qui opère à une fréquence différente, un modèle de similitude en fréquence a été ajouté au modèle de prédiction. La séparation des effets (gaz, nuage, pluie) est réalisée par un réseau de neurones, puis chaque composante est transposée à la fréquence voulue grâce à des coefficients de similitude spécifiques. L'incertitude sur ces coefficients de similitude implique une gestion combinée des erreurs de prédiction et des erreurs dues à la similitude.
Le modèle de prédiction avec similitude en fréquence a été développé avec des mesures de l'affaiblissement des balises 20/30 GHz du satellite OLYMPUS et est ensuite testé avec des données récentes de l'expérience SYRACUSE3 20/44 GHz. Les premiers résultats de cette expérience concernant les statistiques à long terme de l'affaiblissement sont ensuite présentés et comparés aux modèles standard de l'ITU.
Afin de mieux comprendre le lien entre l'affaiblissement et ses causes physiques, une approche basée sur les propriétés multifractales de la pluie est ensuite présentée. En effet, l'analogie entre la pluie et la finance peut être étendue, car ces deux phénomènes sont liés à des processus turbulents et possèdent des propriétés d'invariance d'échelle intéressantes. Malheureusement, l'analyse multifractale directe des séries temporelles d'affaiblissement ne donne pas de résultats satisfaisants. L'analyse multifractale est donc appliquée dans un premier temps à des séries temporelles de taux de pluie. Une évaluation de l'effet de l'intermittence pluie-non pluie sur l'analyse multifractale montre qu'elle provoque une cassure des relations d'invariance d'échelle et peut biaiser considérablement l'estimation des paramètres. L'analyse multifractale est alors réalisée évènement par évènement, c'est-à-dire avec des séries temporelles ininterrompues. Les résultats montrent que la pluie peut être modélisée par un FIF (Fractionally Integrated Flux) auquel on applique un seuil afin de reproduire l'intermittence pluie-non pluie.
La modélisation multifractale de la pluie est ensuite utilisée afin de simuler une liaison Terre-Satellite et de générer des séries synthétiques d'affaiblissement par la pluie. L'analyse de ces séries simulées permet de mieux comprendre pourquoi l'affaiblissement est difficile à modéliser. En particulier, bien que le champ pluie soit multifractal, les séries temporelles d'affaiblissement ne présentent pas de propriétés d'invariance d'échelle stables et peuvent même présenter un redressement du spectre de puissance aux hautes fréquences. Ces résultats montrent que le redressement du spectre observé empiriquement n'est pas dû uniquement au bruit de scintillation.
Gagné, Christian. "Prévisions robustes pour séries temporelles multivariées." Thesis, 2007. http://hdl.handle.net/1866/18185.
Saint-Frard, Robinson. "Sur la validation des modèles de séries chronologiques spatio-temporelles multivariées." Thèse, 2011. http://hdl.handle.net/1866/5331.
In this master thesis, time series models are studied, which have also a spatial component, in addition to the usual time index. More particularly, we study a certain class of models, the Generalized Space-Time AutoRegressive (GSTAR) time series models. First, links are considered between Vector AutoRegressive models(VAR) and GSTAR models. We obtain explicitly the asymptotic distribution of the residual autocovariances for the GSTAR models, assuming that the error term is a Gaussian white noise, which is a first original contribution. From that result, test statistics of the portmanteau type are proposed, and their asymptotic distributions are studied. In order to illustrate the behaviour of the test statistics, a simulation study is conducted where GSTAR models are simulated and correctly fitted. The methodology is illustrated with monthly real data concerning the production of tea in west Java for 24 cities from the period January 1992 to December 1999.
Dans ce mémoire, nous avons utilisé le logiciel R pour la programmation.
Moutacalli, Mohamed Tarik. "Prédiction et reconnaissance d'activités dans un habitat intelligent basées sur les séries temporelles et la fouille de données temporelles." Thèse, 2015. http://constellation.uqac.ca/3748/1/Moutacalli_uqac_0862D_10178.pdf.
"Méthodes non linéaires pour séries temporelles : prédiction par Double Quantification Vectorielle et sélection du délai en hautes dimensions." Université catholique de Louvain, 2007. http://edoc.bib.ucl.ac.be:81/ETD-db/collection/available/BelnUcetd-06142007-105431/.