Amblard, PierreOlivier
Overview
Works:  22 works in 30 publications in 2 languages and 147 library holdings 

Roles:  Thesis advisor, Other, Author, Contributor, Opponent, Editor 
Publication Timeline
.
Most widely held works by
PierreOlivier Amblard
Statistiques d'ordre supérieur pour le traitement du signal by
JeanLouis Lacoume(
Book
)
2 editions published in 1997 in French and held by 66 WorldCat member libraries worldwide
2 editions published in 1997 in French and held by 66 WorldCat member libraries worldwide
A Primer on Reproducing Kernel Hilbert Spaces by
Jonathan H Manton(
)
5 editions published between 2014 and 2015 in English and Undetermined and held by 46 WorldCat member libraries worldwide
Reproducing kernel Hilbert spaces are elucidated without assuming prior familiarity with Hilbert spaces. Compared with extant pedagogic material, greater care is placed on motivating the definition of reproducing kernel Hilbert spaces and explaining when and why these spaces are efficacious. The novel viewpoint is that reproducing kernel Hilbert space theory studies extrinsic geometry, associating with each geometric configuration a canonical overdetermined coordinate system. This coordinate system varies continuously with changing geometric configurations, making it wellsuited for studying problems whose solutions also vary continuously with changing geometry. This primer can also serve as an introduction to infinitedimensional linear algebra because reproducing kernel Hilbert spaces have more properties in common with Euclidean spaces than do more general Hilbert spaces
5 editions published between 2014 and 2015 in English and Undetermined and held by 46 WorldCat member libraries worldwide
Reproducing kernel Hilbert spaces are elucidated without assuming prior familiarity with Hilbert spaces. Compared with extant pedagogic material, greater care is placed on motivating the definition of reproducing kernel Hilbert spaces and explaining when and why these spaces are efficacious. The novel viewpoint is that reproducing kernel Hilbert space theory studies extrinsic geometry, associating with each geometric configuration a canonical overdetermined coordinate system. This coordinate system varies continuously with changing geometric configurations, making it wellsuited for studying problems whose solutions also vary continuously with changing geometry. This primer can also serve as an introduction to infinitedimensional linear algebra because reproducing kernel Hilbert spaces have more properties in common with Euclidean spaces than do more general Hilbert spaces
Influence des fluctuations sur l'échantillonnage et la quantification dans le système visuel by
Cédric Duchêne(
)
2 editions published in 2007 in French and held by 3 WorldCat member libraries worldwide
Recent studies have shown that biologie al neural systems are able to use noise and non linearities to improve the information processing which is occurred in. This thesis focus on this topic. We investigate the links between some operations of signal processing whose potentially appear in the visual system and its internaI noises. A description of the human visual system and its different sources of noise is done in first chapter. We study in the second part the link between the irregular retinal sampling and the random fixational eye movements. We use a simple model of retina. For sever al kind of fluctuations we show that the likeness of the image projected on the model of retina and the real scene can be improved by random movements. ln the third chapter we are interested in a problem which is recurrent in biological systems such as the visual system: noisy binary detection tasks. The influence of the internaI noise of the first layers of neurons of the visual system on the performance of the detection tasks is characterized. To simulate the internaI noise observed in biological neural networks we propose to use stochastic quantizers. A stochastic quantizer is a quantizer whitch of thresholds are perturbed randomly by threshold noises. Once again we observe that the threshold noise can improve the detection performance by decreasing the probability of error
2 editions published in 2007 in French and held by 3 WorldCat member libraries worldwide
Recent studies have shown that biologie al neural systems are able to use noise and non linearities to improve the information processing which is occurred in. This thesis focus on this topic. We investigate the links between some operations of signal processing whose potentially appear in the visual system and its internaI noises. A description of the human visual system and its different sources of noise is done in first chapter. We study in the second part the link between the irregular retinal sampling and the random fixational eye movements. We use a simple model of retina. For sever al kind of fluctuations we show that the likeness of the image projected on the model of retina and the real scene can be improved by random movements. ln the third chapter we are interested in a problem which is recurrent in biological systems such as the visual system: noisy binary detection tasks. The influence of the internaI noise of the first layers of neurons of the visual system on the performance of the detection tasks is characterized. To simulate the internaI noise observed in biological neural networks we propose to use stochastic quantizers. A stochastic quantizer is a quantizer whitch of thresholds are perturbed randomly by threshold noises. Once again we observe that the threshold noise can improve the detection performance by decreasing the probability of error
Génération de signaux multifractals possédant une structure de branchement sousjacente by
Geoffrey Decrouez(
)
2 editions published in 2009 in English and held by 3 WorldCat member libraries worldwide
La géométrie fractale, développée par Mandelbrot dans les années 70, a connu un essor considérable ces 20 dernières années. Dans cette thèse, je m'intéresse à la génération de signaux dits fractals et multifractals. J'étudie en particulier 2 modèles, dont leur point commun est leur structure d'arbre de branchement sous jacente. Le premier modèle est une généralisation des Systèmes de fonctions Itérés ou IFS, introduits par Hutchinson dans les années 80. Les IFS constituent un moyen simple et efficace pour produire des ensembles et des processus fractals en itérant un nombre fixed d'opérateurs. L'idée est d'autoriser un nombre aléatoire d'opérateurs aléatoires à chaque itération de l'algorithme. Nous donnons des conditions simples et faciles à vérifier sous lesquelles l'IFS admet un point fixe. Quelques propriétés du point fixe sont également étudiées. Le deuxième modèle, que nous appellons Multifractal Embedded Branching Process (MEBP), s'obtient à l'aide d'un changement de temps multifractal d'un processus à invariance d'échelle discrète, le processus EBP Canonique (CEBP). Nous donnons un algorithm efficace de simulation "online" de ces processus, permettant de générer X(n + 1) à partir de X(n) en O(log n) opérations. Nous obtenons également un borne supérieure pour le spectre multifractal du changement de temps et confirmons les résultats théoriques à l'aide de simulations. Les mouvements Browniens en temps multifractal sont des cas particuliers des processus MEBP, ce qui suggère une application potentielle des processus MEBP en finance. Enfin, nous proposons d'imiter un mouvement Brownien fractionnaire à l'aide d'un processus MEBP
2 editions published in 2009 in English and held by 3 WorldCat member libraries worldwide
La géométrie fractale, développée par Mandelbrot dans les années 70, a connu un essor considérable ces 20 dernières années. Dans cette thèse, je m'intéresse à la génération de signaux dits fractals et multifractals. J'étudie en particulier 2 modèles, dont leur point commun est leur structure d'arbre de branchement sous jacente. Le premier modèle est une généralisation des Systèmes de fonctions Itérés ou IFS, introduits par Hutchinson dans les années 80. Les IFS constituent un moyen simple et efficace pour produire des ensembles et des processus fractals en itérant un nombre fixed d'opérateurs. L'idée est d'autoriser un nombre aléatoire d'opérateurs aléatoires à chaque itération de l'algorithme. Nous donnons des conditions simples et faciles à vérifier sous lesquelles l'IFS admet un point fixe. Quelques propriétés du point fixe sont également étudiées. Le deuxième modèle, que nous appellons Multifractal Embedded Branching Process (MEBP), s'obtient à l'aide d'un changement de temps multifractal d'un processus à invariance d'échelle discrète, le processus EBP Canonique (CEBP). Nous donnons un algorithm efficace de simulation "online" de ces processus, permettant de générer X(n + 1) à partir de X(n) en O(log n) opérations. Nous obtenons également un borne supérieure pour le spectre multifractal du changement de temps et confirmons les résultats théoriques à l'aide de simulations. Les mouvements Browniens en temps multifractal sont des cas particuliers des processus MEBP, ce qui suggère une application potentielle des processus MEBP en finance. Enfin, nous proposons d'imiter un mouvement Brownien fractionnaire à l'aide d'un processus MEBP
Statistiques d'ordre supérieur pour les signaux non gaussiens, non linéaires, non stationnaires by
PierreOlivier Amblard(
)
1 edition published in 1994 in French and held by 2 WorldCat member libraries worldwide
DURANT LES TRENTE DERNIERES ANNEES, TROIS HYPOTHESES FONDAMENTALES GERAIENT LES THEORIES DEVELOPPEES EN TRAITEMENT DU SIGNAL: GAUSSIANNITE, LINEARITE ET STATIONNARITE. LE BIEN FONDE DE CES PROPRIETES EST VERIFIE DANS BON NOMBRE D'APPLICATIONS REELLES. TOUTEFOIS, CERTAINES SITUATIONS NE PEUVENT ETRE ETUDIEES EN UTILISANT CES PROPRIETES, ET TOUTES OU PARTIE DOIVENT ETRE ECARTEES. CE MEMOIRE A POUR OBJET L'ETUDE D'OUTILS ADAPTES POUR LA DESCRIPTION DES SIGNAUX NON GAUSSIENS, NON LINEAIRES ET/OU NON STATIONNAIRES. DANS UN PREMIER TEMPS, L'HYPOTHESE DE GAUSSIANNITE EST LEVEE, ET LES STATISTIQUES D'ORDRE SUPERIEUR A DEUX POUR LES SIGNAUX NON GAUSSIENS A VALEURS COMPLEXES SONT INTRODUITES. NOUS ETUDIONS TOUT PARTICULIEREMENT L'INFLUENCE DE LA STATIONNARITE SUR CES OUTILS POUR ARRIVER AUX DEFINITIONS DES MULTICORRELATIONS ET MULTISPECTRES. COMME NONGAUSSIANNITE ET NONLINEARITE SONT LIEES, NOUS ETUDIONS ENSUITE UNE CLASSE PARTICULIERE DE SYSTEMES NON LINEAIRES: LES FILTRES DE VOLTERRA. LEURS DEFINITIONS EN TEMPS ET FREQUENCE CONTINUS SONT RAPPELEES AVANT DE PRESENTER LA DEFINITION ET L'IMPLANTATION DE LEURS VERSIONS DISCRETES. L'IDENTIFICATION DE CES SYSTEMES EN MOYENNE QUADRATIQUE EST ALORS APPLIQUEE A LA METHODE DE SOUSTRACTION DE BRUIT, ETUDE VALIDEE SUR DES SIGNAUX ISSUS D'UNE EXPERIENCE REELLE. ENFIN, POUR POUVOIR TRAITER DES SIGNAUX NON GAUSSIENS NON STATIONNAIRES, NOUS PRESENTONS LA THEORIE DES REPRESENTATIONS TEMPSFREQUENCE D'ORDRE SUPERIEUR A DEUX. CETTE THEORIE, DEVELOPPEE D'UNE FACON DEDUCTIVE POUR LES SIGNAUX DETERMINISTES, EST ETENDUE AUX SIGNAUX ALEATOIRES. DES DISCUSSIONS SUR LA COMPLEXITE DES OUTILS OBTENUS SONT MENEES, ET UNE APPLICATION A LA DETECTION DE SIGNAUX TRANSITOIRES EST PROPOSEE
1 edition published in 1994 in French and held by 2 WorldCat member libraries worldwide
DURANT LES TRENTE DERNIERES ANNEES, TROIS HYPOTHESES FONDAMENTALES GERAIENT LES THEORIES DEVELOPPEES EN TRAITEMENT DU SIGNAL: GAUSSIANNITE, LINEARITE ET STATIONNARITE. LE BIEN FONDE DE CES PROPRIETES EST VERIFIE DANS BON NOMBRE D'APPLICATIONS REELLES. TOUTEFOIS, CERTAINES SITUATIONS NE PEUVENT ETRE ETUDIEES EN UTILISANT CES PROPRIETES, ET TOUTES OU PARTIE DOIVENT ETRE ECARTEES. CE MEMOIRE A POUR OBJET L'ETUDE D'OUTILS ADAPTES POUR LA DESCRIPTION DES SIGNAUX NON GAUSSIENS, NON LINEAIRES ET/OU NON STATIONNAIRES. DANS UN PREMIER TEMPS, L'HYPOTHESE DE GAUSSIANNITE EST LEVEE, ET LES STATISTIQUES D'ORDRE SUPERIEUR A DEUX POUR LES SIGNAUX NON GAUSSIENS A VALEURS COMPLEXES SONT INTRODUITES. NOUS ETUDIONS TOUT PARTICULIEREMENT L'INFLUENCE DE LA STATIONNARITE SUR CES OUTILS POUR ARRIVER AUX DEFINITIONS DES MULTICORRELATIONS ET MULTISPECTRES. COMME NONGAUSSIANNITE ET NONLINEARITE SONT LIEES, NOUS ETUDIONS ENSUITE UNE CLASSE PARTICULIERE DE SYSTEMES NON LINEAIRES: LES FILTRES DE VOLTERRA. LEURS DEFINITIONS EN TEMPS ET FREQUENCE CONTINUS SONT RAPPELEES AVANT DE PRESENTER LA DEFINITION ET L'IMPLANTATION DE LEURS VERSIONS DISCRETES. L'IDENTIFICATION DE CES SYSTEMES EN MOYENNE QUADRATIQUE EST ALORS APPLIQUEE A LA METHODE DE SOUSTRACTION DE BRUIT, ETUDE VALIDEE SUR DES SIGNAUX ISSUS D'UNE EXPERIENCE REELLE. ENFIN, POUR POUVOIR TRAITER DES SIGNAUX NON GAUSSIENS NON STATIONNAIRES, NOUS PRESENTONS LA THEORIE DES REPRESENTATIONS TEMPSFREQUENCE D'ORDRE SUPERIEUR A DEUX. CETTE THEORIE, DEVELOPPEE D'UNE FACON DEDUCTIVE POUR LES SIGNAUX DETERMINISTES, EST ETENDUE AUX SIGNAUX ALEATOIRES. DES DISCUSSIONS SUR LA COMPLEXITE DES OUTILS OBTENUS SONT MENEES, ET UNE APPLICATION A LA DETECTION DE SIGNAUX TRANSITOIRES EST PROPOSEE
On directed information theory and Granger causality graphs by
PierreOlivier Amblard(
)
1 edition published in 2010 in English and held by 2 WorldCat member libraries worldwide
1 edition published in 2010 in English and held by 2 WorldCat member libraries worldwide
Traitement du signal dans le domaine compressé et quantification sur un bit : deux outils pour les contextes sous contraintes
de communication by
Augusto Zebadúa(
)
1 edition published in 2017 in French and held by 2 WorldCat member libraries worldwide
Monitoring physical phenomena by using a network of sensors (autonomous but interconnected) is highly constrained in energy consumption, mainly for data transmission. In this context, this thesis proposes signal processing tools to reduce communications without compromising computational accuracy in subsequent calculations. The complexity of these methods is reduced, so as to consume only little additional energy. Our two building blocks are compression during signal acquisition (Compressive Sensing) and CoarseQuantization (1 bit). We first study the Compressed Correlator, an estimator which allows for evaluating correlation functions, timedelay, and spectral densities directly from compressed signals. Its performance is compared with the usual correlator. As we show, if the signal of interest has limited frequency content, the proposed estimator significantly outperforms theconventional correlator. Then, inspired by the coarse quantization correlators from the 50s and 60s, two new correlators are studied: The 1bit Compressed and the Hybrid Compressed, which can also outperform their uncompressed counterparts. Finally, we show the applicability of these methods in the context of interest through the exploitation of real data
1 edition published in 2017 in French and held by 2 WorldCat member libraries worldwide
Monitoring physical phenomena by using a network of sensors (autonomous but interconnected) is highly constrained in energy consumption, mainly for data transmission. In this context, this thesis proposes signal processing tools to reduce communications without compromising computational accuracy in subsequent calculations. The complexity of these methods is reduced, so as to consume only little additional energy. Our two building blocks are compression during signal acquisition (Compressive Sensing) and CoarseQuantization (1 bit). We first study the Compressed Correlator, an estimator which allows for evaluating correlation functions, timedelay, and spectral densities directly from compressed signals. Its performance is compared with the usual correlator. As we show, if the signal of interest has limited frequency content, the proposed estimator significantly outperforms theconventional correlator. Then, inspired by the coarse quantization correlators from the 50s and 60s, two new correlators are studied: The 1bit Compressed and the Hybrid Compressed, which can also outperform their uncompressed counterparts. Finally, we show the applicability of these methods in the context of interest through the exploitation of real data
Solar EUV/FUV irradiance variations: analysis and observational strategy by
Matthieu Kretzschmar(
)
1 edition published in 2009 in English and held by 2 WorldCat member libraries worldwide
1 edition published in 2009 in English and held by 2 WorldCat member libraries worldwide
Méthodes d'illumination et de détection innovantes pour l'amélioration du contraste et de la résolution en imagerie moléculaire
de fluorescence en rétrodiffusion by
Frédéric Fantoni(
)
1 edition published in 2014 in French and held by 2 WorldCat member libraries worldwide
Intraoperative fluorescence imaging in reflectance geometry is an attractive imaging modality to noninvasively monitor fluorescencetargeted tumors. However, in some situations, this kind of imaging suffers from a lack of depth penetration and a poor resolution due to the diffusive nature of photons in tissue. The objective of the thesis was to tackle these limitations. Rather than using a widefield illumination like usual systems, the technique developed relies on the scanning of the medium with a laser line illumination and the acquisition of images at each position of excitation. Several detection schemes are proposed to take advantage of the stack of images acquired to enhance the resolution and the contrast of the final image. These detection techniques were tested both in simulation with the NIRFAST software and a MonteCarlo algorithm and experimentally. The experimental validation was performed on tissuelike phantoms and in vivo with a preliminary testing. The results are compared to those obtained with a classical widefield illumination. As they enhance both the contrast and the resolution, these methods allow us to image deeper targets by reducing the negative effects of parasite signals and diffusion
1 edition published in 2014 in French and held by 2 WorldCat member libraries worldwide
Intraoperative fluorescence imaging in reflectance geometry is an attractive imaging modality to noninvasively monitor fluorescencetargeted tumors. However, in some situations, this kind of imaging suffers from a lack of depth penetration and a poor resolution due to the diffusive nature of photons in tissue. The objective of the thesis was to tackle these limitations. Rather than using a widefield illumination like usual systems, the technique developed relies on the scanning of the medium with a laser line illumination and the acquisition of images at each position of excitation. Several detection schemes are proposed to take advantage of the stack of images acquired to enhance the resolution and the contrast of the final image. These detection techniques were tested both in simulation with the NIRFAST software and a MonteCarlo algorithm and experimentally. The experimental validation was performed on tissuelike phantoms and in vivo with a preliminary testing. The results are compared to those obtained with a classical widefield illumination. As they enhance both the contrast and the resolution, these methods allow us to image deeper targets by reducing the negative effects of parasite signals and diffusion
Identification passive en acoustique : estimateurs et applications au SHM by
Rémy Vincent(
)
1 edition published in 2016 in French and held by 2 WorldCat member libraries worldwide
Ward identity is a relationship that enables damped linear system identification, ie the estimation its caracteristic properties. This identity is used to provide new observation models that are available in an estimation context where sources are uncontrolled by the user. An estimation and detection theory is derived from these models and various performances studies areconducted for several estimators. The reach of the proposed methods is extended to Structural Health Monitoring (SHM), that aims at measuring and tracking the health of buildings, such as a bridge or a skyscraper for instance. The acoustic modality is chosen as it provides complementary parameters estimation to the state of the art in SHM, such as structural and geometrical parameters recovery. Some scenarios are experimentally illustrated by using the developed algorithms, adapted to fit the constrains set by embedded computation on anautonomous sensor network
1 edition published in 2016 in French and held by 2 WorldCat member libraries worldwide
Ward identity is a relationship that enables damped linear system identification, ie the estimation its caracteristic properties. This identity is used to provide new observation models that are available in an estimation context where sources are uncontrolled by the user. An estimation and detection theory is derived from these models and various performances studies areconducted for several estimators. The reach of the proposed methods is extended to Structural Health Monitoring (SHM), that aims at measuring and tracking the health of buildings, such as a bridge or a skyscraper for instance. The acoustic modality is chosen as it provides complementary parameters estimation to the state of the art in SHM, such as structural and geometrical parameters recovery. Some scenarios are experimentally illustrated by using the developed algorithms, adapted to fit the constrains set by embedded computation on anautonomous sensor network
Fusion de données inertielles et magnétiques pour l'estimation de l'attitude sous contrainte énergétique d'un corps rigide
accéléré by
Aida Makni(
)
1 edition published in 2016 in French and held by 2 WorldCat member libraries worldwide
In this PhD. thesis we deal with attitude estimation of accelerated rigid body moving in the 3D space using quaternion parameterization. This problem has been widely studied in the literature in various application areas. The main objective of the thesis is to propose new methods for data fusion to combine inertial gyros) and magnetic measurements. The first challenge concerns the attitude estimation during dynamic cases, in which external acceleration of the body is not negligible compared to the Gravity. Two main approaches are proposed in this context. Firstly, a quatenionbased adaptive Kalman filter (qAKF) was designed in order to compensate for such external acceleration. Precisely, a smart detector is designed to decide whether the body is in static or dynamic case. Then, the covariance matrix of the external acceleration is estimated to tune the filter gain. Second, we developed descriptor filter based on a new formulation of the dynamic model where the process model is fed by accelerometer measurements while observation model is fed by gyros and magnetometer measurements. Such modeling gives rise to a descriptor system. The resulting model allows taking the external acceleration of the body into account in a very efficient way. The second challenge is related to the energy consumption issue of gyroscope, considered as the most power consuming sensor. We study the way to reduce the gyro measurements acquisition by switching on/off the sensor while maintaining an acceptable attitude estimation. The effciency of the proposed methods is evaluated by means of numerical simulations and experimental tests
1 edition published in 2016 in French and held by 2 WorldCat member libraries worldwide
In this PhD. thesis we deal with attitude estimation of accelerated rigid body moving in the 3D space using quaternion parameterization. This problem has been widely studied in the literature in various application areas. The main objective of the thesis is to propose new methods for data fusion to combine inertial gyros) and magnetic measurements. The first challenge concerns the attitude estimation during dynamic cases, in which external acceleration of the body is not negligible compared to the Gravity. Two main approaches are proposed in this context. Firstly, a quatenionbased adaptive Kalman filter (qAKF) was designed in order to compensate for such external acceleration. Precisely, a smart detector is designed to decide whether the body is in static or dynamic case. Then, the covariance matrix of the external acceleration is estimated to tune the filter gain. Second, we developed descriptor filter based on a new formulation of the dynamic model where the process model is fed by accelerometer measurements while observation model is fed by gyros and magnetometer measurements. Such modeling gives rise to a descriptor system. The resulting model allows taking the external acceleration of the body into account in a very efficient way. The second challenge is related to the energy consumption issue of gyroscope, considered as the most power consuming sensor. We study the way to reduce the gyro measurements acquisition by switching on/off the sensor while maintaining an acceptable attitude estimation. The effciency of the proposed methods is evaluated by means of numerical simulations and experimental tests
Particle Filtering Equalization Method for a Satellite Communication Channel by
Stéphane Sénécal(
)
1 edition published in 2004 in English and held by 2 WorldCat member libraries worldwide
1 edition published in 2004 in English and held by 2 WorldCat member libraries worldwide
Modèle d'interaction et performances du traitement du signal multimodal by
Saloua Chlaily(
)
1 edition published in 2018 in French and held by 2 WorldCat member libraries worldwide
The joint processing of multimodal measurements is supposed to lead to better performances than those obtained using a single modality or several modalities independently. However, in literature, there are examples that show that is not always true. In this thesis, we analyze, in terms of mutual information and estimation error, the different situations of multimodal analysis in order to determine the conditions to achieve the optimal performances.In the first part, we consider the simple case of two or three modalities, each associated with noisy measurement of a signal. These modalities are linked through the correlations between the useful parts of the signal and the correlations between the noises. We show that the performances are improved if the links between the modalities are exploited. In the second part, we study the impact on performance of wrong links between modalities. We show that these false assumptions decline the performance, which can become lower than the performance achieved using a single modality.In the general case, we model the multiple modalities as a noisy Gaussian channel. We then extend literature results by considering the impact of the errors on signal and noise probability densities on the information transmitted by the channel. We then analyze this relationship in the case of a simple model of two modalities. Our results show in particular the unexpected fact that a double mismatch of the noise and the signal can sometimes compensate for each other, and thus lead to very good performances
1 edition published in 2018 in French and held by 2 WorldCat member libraries worldwide
The joint processing of multimodal measurements is supposed to lead to better performances than those obtained using a single modality or several modalities independently. However, in literature, there are examples that show that is not always true. In this thesis, we analyze, in terms of mutual information and estimation error, the different situations of multimodal analysis in order to determine the conditions to achieve the optimal performances.In the first part, we consider the simple case of two or three modalities, each associated with noisy measurement of a signal. These modalities are linked through the correlations between the useful parts of the signal and the correlations between the noises. We show that the performances are improved if the links between the modalities are exploited. In the second part, we study the impact on performance of wrong links between modalities. We show that these false assumptions decline the performance, which can become lower than the performance achieved using a single modality.In the general case, we model the multiple modalities as a noisy Gaussian channel. We then extend literature results by considering the impact of the errors on signal and noise probability densities on the information transmitted by the channel. We then analyze this relationship in the case of a simple model of two modalities. Our results show in particular the unexpected fact that a double mismatch of the noise and the signal can sometimes compensate for each other, and thus lead to very good performances
Méthodes de simulation MonteCarlo par chaînes de Markov pour l'estimation de modèles : applications en séparation de
sources et en égalisation by
Stéphane Sénécal(
Book
)
2 editions published in 2002 in French and held by 2 WorldCat member libraries worldwide
This thesis proposes to study and apply Markov chain MonteCarlo (MCMC) simulation methods for solving estimation problems arising in the field of signal processing. Prior hypothesis on signals and/or on transfert function models can be taken into account thanks to a Bayesian approach leading to classical estimators, posterior mean and maximum a posteriori for instance, whose computation is generally difficult. MonteCarlo estimation techniques are thus considered and implemented with Markov chain simulation schemes. In a first step, estimation problems dealing with source separation models are considered. The case of digital communications is specifically focused on with the study of source signals associated to PSK modulations. A separation method based on the Gibbs sampling algorithm is proposed and illustrated with numerical experiments. The algorithm makes it possible to separate underdetermined mixtures and can be modified in order to solve other problems associated to this model: estimation of the number of state for the constellations of source signals, estimation of the number of the source signals. In a second step, an equalisation problem is tackled in the context of satellite communications. The Bayesian approach and its implementation through a MonteCarlo estimation technique make it possible to take into account explicitely the nonlinearity of the amplification stage in the satellite. A sequential simulation method based on particle filtering is then proposed for equalising blindly and robustly the complete transmission chain
2 editions published in 2002 in French and held by 2 WorldCat member libraries worldwide
This thesis proposes to study and apply Markov chain MonteCarlo (MCMC) simulation methods for solving estimation problems arising in the field of signal processing. Prior hypothesis on signals and/or on transfert function models can be taken into account thanks to a Bayesian approach leading to classical estimators, posterior mean and maximum a posteriori for instance, whose computation is generally difficult. MonteCarlo estimation techniques are thus considered and implemented with Markov chain simulation schemes. In a first step, estimation problems dealing with source separation models are considered. The case of digital communications is specifically focused on with the study of source signals associated to PSK modulations. A separation method based on the Gibbs sampling algorithm is proposed and illustrated with numerical experiments. The algorithm makes it possible to separate underdetermined mixtures and can be modified in order to solve other problems associated to this model: estimation of the number of state for the constellations of source signals, estimation of the number of the source signals. In a second step, an equalisation problem is tackled in the context of satellite communications. The Bayesian approach and its implementation through a MonteCarlo estimation technique make it possible to take into account explicitely the nonlinearity of the amplification stage in the satellite. A sequential simulation method based on particle filtering is then proposed for equalising blindly and robustly the complete transmission chain
Blind Equalization of a Nonlinear Satellite System Using MCMC Simulation Methods by
Stéphane Sénécal(
)
1 edition published in 2002 in English and held by 2 WorldCat member libraries worldwide
1 edition published in 2002 in English and held by 2 WorldCat member libraries worldwide
Sur l'échantillonnage des processus ponctuels déterminantaux by
Guillaume, Michel, Jean Gautier(
)
1 edition published in 2020 in English and held by 1 WorldCat member library worldwide
Determinantal point processes (DPPs) generate random configuration of points where the points tend to repel each other. The notion of repulsion is encoded by the subdeterminants of a kernel matrix, in the sense of kernel methods in machine learning. This special algebraic form makes DPPs attractive both in statistical and computational terms. This thesis focuses on sampling from such processes, that is on developing simulation methods for DPPs. Applications include numerical integration, recommender systems or the summarization of a large corpus of data. In the finite setting, we establish the correspondence between sampling from a specific type of DPPs, called projection DPPs, and solving a randomized linear program. In this light, we devise an efficient Markovchainbased sampling method. In the continuous case, some classical DPPs can be sampled by computing the eigenvalues of carefully randomized tridiagonal matrices. We provide an elementary and unifying treatment of such models, from which we derive an approximate sampling method for more general models. In higher dimension, we consider a special class of DPPs used for numerical integration. We implement a tailored version of a known exact sampler, which allows us to compare the properties of Monte Carlo estimators in new regimes. In the context of reproducible research, we develop an opensource Python toolbox, named DPPy, which implements the state of the art sampling methods for DPPs
1 edition published in 2020 in English and held by 1 WorldCat member library worldwide
Determinantal point processes (DPPs) generate random configuration of points where the points tend to repel each other. The notion of repulsion is encoded by the subdeterminants of a kernel matrix, in the sense of kernel methods in machine learning. This special algebraic form makes DPPs attractive both in statistical and computational terms. This thesis focuses on sampling from such processes, that is on developing simulation methods for DPPs. Applications include numerical integration, recommender systems or the summarization of a large corpus of data. In the finite setting, we establish the correspondence between sampling from a specific type of DPPs, called projection DPPs, and solving a randomized linear program. In this light, we devise an efficient Markovchainbased sampling method. In the continuous case, some classical DPPs can be sampled by computing the eigenvalues of carefully randomized tridiagonal matrices. We provide an elementary and unifying treatment of such models, from which we derive an approximate sampling method for more general models. In higher dimension, we consider a special class of DPPs used for numerical integration. We implement a tailored version of a known exact sampler, which allows us to compare the properties of Monte Carlo estimators in new regimes. In the context of reproducible research, we develop an opensource Python toolbox, named DPPy, which implements the state of the art sampling methods for DPPs
Periodic models and variations applied to health problems by
Paulo Roberto Prezotti Filho(
)
1 edition published in 2019 in English and held by 1 WorldCat member library worldwide
Ce manuscrit porte sur certaines extensions à des séries temporelles prenant des valeurs entières du modèle paramétrique périodique autorégressif établi pour des séries prenant des valeurs réelles. Les modèles que nous considérons sont basés sur l'utilisation de l'opérateur de Steutel et Van Harn (1979) et généralisent le processus autorégressif stationnaire à valeurs entières (INAR) introduit par AlOsh & Alzaid (1987) à des séries de comptage périodiquement corrélées. Ces généralisations incluent l'introduction d'un opérateur périodique, la prise en compte d'une structure d'autocorrélation plus complexe dont l'ordre est supérieur à un, l'apparition d'innovations de variances périodiques mais aussi à inflation de zéro par rapport à une loi discrète donnée dans la famille des distributions exponentielles, ainsi que l'utilisation de covariables explicatives. Ces extensions enrichissent considérablement le domaine d'applicabilité des modèles de type INAR. Sur le plan théorique, nous établissons des propriétés mathématiques de nos modèles telles que l'existence, l'unicité, la stationnarité périodique de solutions aux équations définissant les modèles. Nous proposons trois méthodes d'estimation des paramètres des modèles dont une méthode des moments basée sur des équations du type YuleWalker, une méthode des moindres carrés conditionnels, et une méthode du quasi maximum de vraisemblance (QML) basée sur la maximisation d'une vraisemblance gaussienne. Nous établissons la consistance et la normalité asymptotique de ces procédures d'estimation. Des simulations de type Monte Carlo illustrent leur comportement pour différentes tailles finies d'échantillon. Les modèles sont ensuite ajustés à des données réelles et utilisés à des fins de prédiction. La première extension du modèle INAR que nous proposons consiste à introduire deux opérateurs de Steutel et Van Harn périodiques, l'un modélisant les autocorrélations partielles d'ordre un sur chaque période et l'autre captant la saisonnalité périodique des données. Grâce à une représentation vectorielle du processus, nous établissons les conditions l'existence et d'unicité d'une solution périodiquement corrélées aux équations définissant le modèle. Dans le cas où les innovations suivent des lois de Poisson, nous étudions la loi marginale du processus. Á titre d'exemple d'application sur des données réelles, nous ajustons ce modèle à des données de comptage journalières du nombre de personnes ayant reçu des antibiotiques pour le traitement de maladies respiratoires dans la région de Vitória au Brésil. Comme les affections respiratoires sont fortement corrélées au niveau de pollution atmosphérique et aux conditions climatiques, la structure de corrélation des nombres quotidiens de personnes recevant des antibiotiques montre, entre autres caractéristiques, une périodicité et un caractère saisonnier hebdomadaire. Nous étendons ensuite ce modèle à des données présentant des autocorrélations partielles périodiques d'ordre supérieur à un. Nous étudions les propriétés statistiques du modèle, telles que la moyenne, la variance, les distributions marginales et jointes. Nous ajustons ce modèle au nombre quotidien de personnes recevant du service d'urgence de l'hôpital public de Vitória un traitement pour l'asthme. Enfin, notre dernière extension porte sur l'introduction d'innovations suivant une loi de Poisson à inflation de zéro dont les paramètres varient périodiquement, et sur l'ajout de covariables expliquant le logarithme de l'intensité de la loi de Poisson. Nous établissons certaines propriétés statistiques du modèle et nous mettons en oeuvre la méthode du QML pour estimer ses paramètres. Enfin, nous appliquons cette modélisation à des données journalières du nombre de personnes qui se sont rendues dans le service d'urgence d'un hôpital pour des problèmes respiratoires, et nous utilisons comme covariable la concentration de polluant dans la même zone géographique
1 edition published in 2019 in English and held by 1 WorldCat member library worldwide
Ce manuscrit porte sur certaines extensions à des séries temporelles prenant des valeurs entières du modèle paramétrique périodique autorégressif établi pour des séries prenant des valeurs réelles. Les modèles que nous considérons sont basés sur l'utilisation de l'opérateur de Steutel et Van Harn (1979) et généralisent le processus autorégressif stationnaire à valeurs entières (INAR) introduit par AlOsh & Alzaid (1987) à des séries de comptage périodiquement corrélées. Ces généralisations incluent l'introduction d'un opérateur périodique, la prise en compte d'une structure d'autocorrélation plus complexe dont l'ordre est supérieur à un, l'apparition d'innovations de variances périodiques mais aussi à inflation de zéro par rapport à une loi discrète donnée dans la famille des distributions exponentielles, ainsi que l'utilisation de covariables explicatives. Ces extensions enrichissent considérablement le domaine d'applicabilité des modèles de type INAR. Sur le plan théorique, nous établissons des propriétés mathématiques de nos modèles telles que l'existence, l'unicité, la stationnarité périodique de solutions aux équations définissant les modèles. Nous proposons trois méthodes d'estimation des paramètres des modèles dont une méthode des moments basée sur des équations du type YuleWalker, une méthode des moindres carrés conditionnels, et une méthode du quasi maximum de vraisemblance (QML) basée sur la maximisation d'une vraisemblance gaussienne. Nous établissons la consistance et la normalité asymptotique de ces procédures d'estimation. Des simulations de type Monte Carlo illustrent leur comportement pour différentes tailles finies d'échantillon. Les modèles sont ensuite ajustés à des données réelles et utilisés à des fins de prédiction. La première extension du modèle INAR que nous proposons consiste à introduire deux opérateurs de Steutel et Van Harn périodiques, l'un modélisant les autocorrélations partielles d'ordre un sur chaque période et l'autre captant la saisonnalité périodique des données. Grâce à une représentation vectorielle du processus, nous établissons les conditions l'existence et d'unicité d'une solution périodiquement corrélées aux équations définissant le modèle. Dans le cas où les innovations suivent des lois de Poisson, nous étudions la loi marginale du processus. Á titre d'exemple d'application sur des données réelles, nous ajustons ce modèle à des données de comptage journalières du nombre de personnes ayant reçu des antibiotiques pour le traitement de maladies respiratoires dans la région de Vitória au Brésil. Comme les affections respiratoires sont fortement corrélées au niveau de pollution atmosphérique et aux conditions climatiques, la structure de corrélation des nombres quotidiens de personnes recevant des antibiotiques montre, entre autres caractéristiques, une périodicité et un caractère saisonnier hebdomadaire. Nous étendons ensuite ce modèle à des données présentant des autocorrélations partielles périodiques d'ordre supérieur à un. Nous étudions les propriétés statistiques du modèle, telles que la moyenne, la variance, les distributions marginales et jointes. Nous ajustons ce modèle au nombre quotidien de personnes recevant du service d'urgence de l'hôpital public de Vitória un traitement pour l'asthme. Enfin, notre dernière extension porte sur l'introduction d'innovations suivant une loi de Poisson à inflation de zéro dont les paramètres varient périodiquement, et sur l'ajout de covariables expliquant le logarithme de l'intensité de la loi de Poisson. Nous établissons certaines propriétés statistiques du modèle et nous mettons en oeuvre la méthode du QML pour estimer ses paramètres. Enfin, nous appliquons cette modélisation à des données journalières du nombre de personnes qui se sont rendues dans le service d'urgence d'un hôpital pour des problèmes respiratoires, et nous utilisons comme covariable la concentration de polluant dans la même zone géographique
Solar EUV / FUV irradiance variations : analysis and observational strategy(
)
1 edition published in 2009 in English and held by 1 WorldCat member library worldwide
1 edition published in 2009 in English and held by 1 WorldCat member library worldwide
Separation of parameterized and delayed sources : application to spectroscopic and multispectral data by
Hassan Mortada(
)
1 edition published in 2018 in English and held by 1 WorldCat member library worldwide
This work is motivated by photoelectron spectroscopy and the study of galaxy kinematics where data respectively correspond to a temporal sequence of spectra and a multispectral image. The objective is to estimate the characteristics (amplitude, spectral position and shape) of peaks embedded in the spectra, but also their evolution within the data. In the considered applications, this evolution is slow since two neighbor spectra are often very similar: this a priori knowledge that will be taken into account in the developed methods. This inverse problem is approached as a delayed source separation problem where spectra and peaks are respectively associated with mixtures and sources. The stateoftheart methods are inadequate because they suppose the source decorrelation and independence, which is not the case. We take advantage of the source knowledge in order to model them by a parameterized function. We first propose an alternating least squares method: the shape parameters are estimated with the LevenbergMarquardt algorithm, whilst the amplitudes and positions are estimated with an algorithm inspired from Orthogonal Matching Pursuit. A second method introduces a regularization term to consider the delay slow evolution; a new joint sparse approximation algorithm is thus proposed. Finally, a third method constrains the evolution of the amplitudes, positions and shape parameters by Bspline functions to guarantee their slow evolution. The Bspline control points are estimated with a nonlinear least squares algorithm. The results on synthetic and real data show that the proposed methods are more effective than stateoftheart methods and as effective as a Bayesian method which is adapted to the problem. Moreover, the proposed methods are significantly faster
1 edition published in 2018 in English and held by 1 WorldCat member library worldwide
This work is motivated by photoelectron spectroscopy and the study of galaxy kinematics where data respectively correspond to a temporal sequence of spectra and a multispectral image. The objective is to estimate the characteristics (amplitude, spectral position and shape) of peaks embedded in the spectra, but also their evolution within the data. In the considered applications, this evolution is slow since two neighbor spectra are often very similar: this a priori knowledge that will be taken into account in the developed methods. This inverse problem is approached as a delayed source separation problem where spectra and peaks are respectively associated with mixtures and sources. The stateoftheart methods are inadequate because they suppose the source decorrelation and independence, which is not the case. We take advantage of the source knowledge in order to model them by a parameterized function. We first propose an alternating least squares method: the shape parameters are estimated with the LevenbergMarquardt algorithm, whilst the amplitudes and positions are estimated with an algorithm inspired from Orthogonal Matching Pursuit. A second method introduces a regularization term to consider the delay slow evolution; a new joint sparse approximation algorithm is thus proposed. Finally, a third method constrains the evolution of the amplitudes, positions and shape parameters by Bspline functions to guarantee their slow evolution. The Bspline control points are estimated with a nonlinear least squares algorithm. The results on synthetic and real data show that the proposed methods are more effective than stateoftheart methods and as effective as a Bayesian method which is adapted to the problem. Moreover, the proposed methods are significantly faster
Kernel LMS à noyau gaussien : conception, analyse et applications à divers contextes by
Wei Gao(
)
1 edition published in 2015 in English and held by 1 WorldCat member library worldwide
The main objective of this thesis is to derive and analyze the Gaussian kernel leastmeansquare (LMS) algorithm within three frameworks involving single and multiple kernels, realvalued and complexvalued, noncooperative and cooperative distributed learning over networks. This work focuses on the stochastic behavior analysis of these kernel LMS algorithms in the mean and meansquare error sense. All the analyses are validated by numerical simulations. First, we review the basic LMS algorithm, reproducing kernel Hilbert space (RKHS), framework and stateoftheart kernel adaptive filtering algorithms. Then, we study the convergence behavior of the Gaussian kernel LMS in the case where the statistics of the elements of the socalled dictionary only partially match the statistics of the input data. We introduced a modified kernel LMS algorithm based on forwardbackward splitting to deal with ℓ₁norm regularization. The stability of the proposed algorithm is then discussed. After a review of two families of multikernel LMS algorithms, we focus on the convergence behavior of the multipleinput multikernel LMS algorithm. More generally, the characteristics of multikernel LMS algorithms are analyzed theoretically and confirmed by simulation results. Next, the augmented complex kernel LMS algorithm is introduced based on the framework of complex multikernel adaptive filtering. Then, we analyze the convergence behavior of algorithm in the meansquare error sense. Finally, in order to cope with the distributed estimation problems over networks, we derive functional diffusion strategies in RKHS. The stability of the algorithm in the mean sense is analyzed
1 edition published in 2015 in English and held by 1 WorldCat member library worldwide
The main objective of this thesis is to derive and analyze the Gaussian kernel leastmeansquare (LMS) algorithm within three frameworks involving single and multiple kernels, realvalued and complexvalued, noncooperative and cooperative distributed learning over networks. This work focuses on the stochastic behavior analysis of these kernel LMS algorithms in the mean and meansquare error sense. All the analyses are validated by numerical simulations. First, we review the basic LMS algorithm, reproducing kernel Hilbert space (RKHS), framework and stateoftheart kernel adaptive filtering algorithms. Then, we study the convergence behavior of the Gaussian kernel LMS in the case where the statistics of the elements of the socalled dictionary only partially match the statistics of the input data. We introduced a modified kernel LMS algorithm based on forwardbackward splitting to deal with ℓ₁norm regularization. The stability of the proposed algorithm is then discussed. After a review of two families of multikernel LMS algorithms, we focus on the convergence behavior of the multipleinput multikernel LMS algorithm. More generally, the characteristics of multikernel LMS algorithms are analyzed theoretically and confirmed by simulation results. Next, the augmented complex kernel LMS algorithm is introduced based on the framework of complex multikernel adaptive filtering. Then, we analyze the convergence behavior of algorithm in the meansquare error sense. Finally, in order to cope with the distributed estimation problems over networks, we derive functional diffusion strategies in RKHS. The stability of the algorithm in the mean sense is analyzed
more
fewer
Audience Level
0 

1  
Kids  General  Special 
Related Identities
 Lacoume, JeanLouis Author
 Comon, Pierre
 Manton, Jonathan H. Author
 Grenoble Images parole signal automatique Other
 Institut national polytechnique (Grenoble) Degree grantor
 SpringerLink (Online service) Other
 École doctorale électronique, électrotechnique, automatique, traitement du signal (Grenoble) Other
 Communauté d'universités et d'établissements Université Grenoble Alpes Degree grantor
 Michel, Olivier (1963....; auteur en traitement du signal) Opponent Thesis advisor
 Sénécal, Stéphane Author
Associated Subjects