Co-clustering leads to parsimony in data visualisation with a number of parameters dramatically r... more Co-clustering leads to parsimony in data visualisation with a number of parameters dramatically reduced in comparison to the dimensions of the data sample. Herein, we propose a new generalized approach for nonlinear mapping by a re-parameterization of the latent block mixture model. The densities modeling the blocks are in an exponential family such that the Gaussian, Bernoulli and Poisson laws are particular cases. The inference of the parameters is derived from the block expectation-maximization algorithm with a Newton-Raphson procedure at the maximization step. Empirical experiments with textual data validate the interest of our generalized model. Revision Notes Q6. I am afraid that it is also unclear what does scalability have to do this. In what sense a model is more scalable when it is "properly defined"? Authors' response: Thanks for this concern. The idea was once again to recall the possible scalability of co-clustering in comparison with row clustering. The sentence have been removed.
Utilisation des modèles de mélange pour la classification automatique de données ordinales
Http Www Theses Fr, 2006
Dans cette these, on s'interesse a l'analyse de donnees ordinales et plus particulieremen... more Dans cette these, on s'interesse a l'analyse de donnees ordinales et plus particulierement aux methodes de classification automatique, egalement appelees apprentissage non supervise, en presentant deux approches originales. La premiere consiste a postuler que les donnees ordinales sont issues de donnees continues discretisees selon des frontieres inconnues. La seconde consiste quant a elle a exprimer en termes probabilistes la notion d'eloignement induite par l'ordre des modalites a partir d'un modele multinomial. Enfin, nous evaluons experimentalement l'efficacite de ces methodes sur des jeux de donnees simulees et reelles. Finalement, une application de ces methodes est faite sur des donnees issues d'enquetes de satisfaction.
Assessing population exposure using multimedia exposure model and spatial analysis techniques
Assessing population exposure using multimedia exposure model and spatial analysis techniquesKeyw... more Assessing population exposure using multimedia exposure model and spatial analysis techniquesKeywords: Model, multimedia, exposure, indicator Analyzing the relationship between environment and health has become a major preoccupation for public health as evidenced by the emergence of the second French national plan for health and environment. Two main axes have been retained: - To identify and manage geographic areas where hotspot exposures are suspected to generate a potential risk to human health; - To reduce exposure inequalities. The explicit objective of this project is to develop a spatial multimedia exposure model able to detect vulnerable populations and to analyze exposure determinants. A multimedia exposure model was developed by INERIS to assess the transfer of Chemicals Of Concerns (COC) from the environment (air, water, soil, food chain) to human exposure through inhalation and ingestion pathways. This project adds a spatial dimension by linking a GIS (Geographic Information System) to the model. Tools were developed using modeling, spatial analysis and geostatistic methods to build and discretize interesting variables and indicators from different supports and resolutions on a 1 km2 regular grid within France. For example, surface soil concentrations were estimated by developing a kriging method able to integrate surface and point spatial supports. We used distance from river and surface water quality measurements to build a proxy for intake by fish ingestion. Downscaling method was developed to map population density at a fine scale using census and land cover data. Eulerian atmospheric dispersion model (CHIMERE) enabled to describe atmospheric deposition and concentrations. Spatial data were used to implement environmental and population characteristics into the multimedia exposure model. The purpose of this work is to present data treatment and model development based on the example of cadmium. Simulation results enable to identify pollutant sources, determinants of exposure, and vulnerable populations. Environmental and exposure datasets will be integrated in the French coordinated, integrated environment and health (spatial) GIS-based platform to map and analyze environmental inequalities.
Assessing heavy metal exposure from ingestion with multimedia exposure model operating in a GIS environment
The last decade has witnessed an increasing interest in assessing health risks caused by exposure... more The last decade has witnessed an increasing interest in assessing health risks caused by exposure to the contaminants present in soils, air, food and water. In this field, Geographic Information System (GIS) may be used to produce maps of exposure and disease to reveal spatial patterns. The SIGFRIED project aims to spatialize an environmental indicator concerning human health using risk assessment methods and mapping environmental disparities. The approach is performed by a risk screening tool which employs an integrated multimedia exposure model able to manage all the steps of the analysis in a georeferenced framework operating in a GIS environment. In this study, we performed a risk assessment of metal exposure (cadmium, lead, selenium, arsenic, chromium, mercury and nickel) with a resolution of 1 km. The modeling was performed on a region of France (Nord-Pas de Calais) characterized by a heavy industrial past. The exposure pathways considered include ingestion of soil, vegetation, meat, egg, milk, fish and drinking water. Concentrations of metals on local food were calculated using a soil pollutants geodatabase and bioaccumulation food-chain models. We used water quality data and distance from river as a proxy for fish consumption exposure. Exposure scenarios were defined for different reference groups (age groups, anatomical and dietary properties, fraction of food produced locally). The results showed that the 2 to 7 age group is always the most vulnerable class of receptor. A spatial pattern analysis was performed to highlight hot-spot areas with significantly elevated risk values. In these areas, arsenic (up to 43 % among metals) and vegetables (up to 31 % among considered pathways) are the principal determinants of the calculated risk. In a next step, the same methodology will be use, adding inhalation pathway and organic pollutants, to map environmental inequalities and to analyse geographic cancer disparities in France.
Caracterisation topologique d'un jeu de donnes images avec les nombres de Betti et un modele generatif
Dans cet article, nous proposons un mod�ele g�en�eratif qui permet d'extraire les nombres de ... more Dans cet article, nous proposons un mod�ele g�en�eratif qui permet d'extraire les nombres de Betti d'un ensemble de vari�et�es de RD �a partir d'un �echantillon. Ce mod�ele est bas�e sur le Complexe Simplicial G�en�eratif, un mod�ele de m�elange dont les composantes sont des simplexes g�eom�etriques convolu�es �a une distribution gaussienne multivari�ee. La nouveaut�e de la m�ethode consiste �a optimiser BIC, un crit�ere statistique de vraisemblance p�enalis�ee, pour obtenir une estimation des invariants topologiques des vari�et�es g�en�eratrices des donn�ees, les nombres de Betti. Les r�esultats d'une telle m�ethode montrent une plus grande robustesse au bruit que ceux d'une m�ethode fond�ee sur une approche purement g�eom�etrique. Le Complexe Simplicial G�en�eratif est compar�e au Witness Complex (WitC) et la persistance homologique sur des donn�ees jouet (sph�ere, tore, bouteille de Klein) et un jeu de donn�ees r�eelles, COIL- 100, une base d'images d'...
A mixture of Kalman filters for online monitoring of railway switches
Assessing the operating state of the railway infrastructure and rolling stock using condition mea... more Assessing the operating state of the railway infrastructure and rolling stock using condition measurements acquired through embedded sensors has become a powerful decision-making support for preventive maintenance strategies. This article introduces a dynamic approach for the online monitoring of railway switch operations. The method is based on modeling the power consumption curves acquired during successive switch operations using conjointly five polynomial regression models whose coefficients are dynamically estimated across a sequence of curves. The experimental study conducted on two real power consumption curve sequences from the French high speed network has shown encouraging results in terms of characterization of the temporal evolution of railway switch operations.
Modèle de mélange poissonien pour l'analyse de données de déplacements
ABSTRACT Studies based on human mobility, including Bicycle Sharing System (BSS) trac analysis, h... more ABSTRACT Studies based on human mobility, including Bicycle Sharing System (BSS) trac analysis, has expanded over the past few years. They give insight of the underlying urban phenomena linked to city dynamics. This paper presents a generative count-series model using Poisson mixtures to automatically analyse and find temporal-based partitions over the Vélib' origin-destination (OD) flow-data. Such an approach may provide latent factors that reveal how regions of diffrent usage interact over the time.
Un graphe génératif pour la classification semi-supervisée
Ingénierie des systèmes d'information, 2010
We introduce a new semi-supervised algorithm based on a generative model. This model combines a G... more We introduce a new semi-supervised algorithm based on a generative model. This model combines a Gaussian mixture model and a generative graph built on the components of this mixture. The combination corresponds to refit the class membership of the mixture component with a propagation process. Both models can be optimized under the maximum likelihood framework and the only hyper-parameter (number of components of the mixture) can be selected with a statistical criterion. Experimental results show that we achieve accuracies comparable to those of rival state-of-the-art algorithms when few labeled data are available. Moreover, it offers the advantage of defining an objective statistical criterion for tuning its parameters, cancelling the need for arbitrary hand-tuning.
Co-clustering leads to parsimony in data visualisation with a number of parameters dramatically r... more Co-clustering leads to parsimony in data visualisation with a number of parameters dramatically reduced in comparison to the dimensions of the data sample. Herein, we propose a new generalized approach for nonlinear mapping by a re-parameterization of the latent block mixture model. The densities modeling the blocks are in an exponential family such that the Gaussian, Bernoulli and Poisson laws are particular cases. The inference of the parameters is derived from the block expectation-maximization algorithm with a Newton-Raphson procedure at the maximization step. Empirical experiments with textual data validate the interest of our generalized model. Revision Notes Q6. I am afraid that it is also unclear what does scalability have to do this. In what sense a model is more scalable when it is "properly defined"? Authors' response: Thanks for this concern. The idea was once again to recall the possible scalability of co-clustering in comparison with row clustering. The sentence have been removed.
Utilisation des modèles de mélange pour la classification automatique de données ordinales
Http Www Theses Fr, 2006
Dans cette these, on s'interesse a l'analyse de donnees ordinales et plus particulieremen... more Dans cette these, on s'interesse a l'analyse de donnees ordinales et plus particulierement aux methodes de classification automatique, egalement appelees apprentissage non supervise, en presentant deux approches originales. La premiere consiste a postuler que les donnees ordinales sont issues de donnees continues discretisees selon des frontieres inconnues. La seconde consiste quant a elle a exprimer en termes probabilistes la notion d'eloignement induite par l'ordre des modalites a partir d'un modele multinomial. Enfin, nous evaluons experimentalement l'efficacite de ces methodes sur des jeux de donnees simulees et reelles. Finalement, une application de ces methodes est faite sur des donnees issues d'enquetes de satisfaction.
Assessing population exposure using multimedia exposure model and spatial analysis techniques
Assessing population exposure using multimedia exposure model and spatial analysis techniquesKeyw... more Assessing population exposure using multimedia exposure model and spatial analysis techniquesKeywords: Model, multimedia, exposure, indicator Analyzing the relationship between environment and health has become a major preoccupation for public health as evidenced by the emergence of the second French national plan for health and environment. Two main axes have been retained: - To identify and manage geographic areas where hotspot exposures are suspected to generate a potential risk to human health; - To reduce exposure inequalities. The explicit objective of this project is to develop a spatial multimedia exposure model able to detect vulnerable populations and to analyze exposure determinants. A multimedia exposure model was developed by INERIS to assess the transfer of Chemicals Of Concerns (COC) from the environment (air, water, soil, food chain) to human exposure through inhalation and ingestion pathways. This project adds a spatial dimension by linking a GIS (Geographic Information System) to the model. Tools were developed using modeling, spatial analysis and geostatistic methods to build and discretize interesting variables and indicators from different supports and resolutions on a 1 km2 regular grid within France. For example, surface soil concentrations were estimated by developing a kriging method able to integrate surface and point spatial supports. We used distance from river and surface water quality measurements to build a proxy for intake by fish ingestion. Downscaling method was developed to map population density at a fine scale using census and land cover data. Eulerian atmospheric dispersion model (CHIMERE) enabled to describe atmospheric deposition and concentrations. Spatial data were used to implement environmental and population characteristics into the multimedia exposure model. The purpose of this work is to present data treatment and model development based on the example of cadmium. Simulation results enable to identify pollutant sources, determinants of exposure, and vulnerable populations. Environmental and exposure datasets will be integrated in the French coordinated, integrated environment and health (spatial) GIS-based platform to map and analyze environmental inequalities.
Assessing heavy metal exposure from ingestion with multimedia exposure model operating in a GIS environment
The last decade has witnessed an increasing interest in assessing health risks caused by exposure... more The last decade has witnessed an increasing interest in assessing health risks caused by exposure to the contaminants present in soils, air, food and water. In this field, Geographic Information System (GIS) may be used to produce maps of exposure and disease to reveal spatial patterns. The SIGFRIED project aims to spatialize an environmental indicator concerning human health using risk assessment methods and mapping environmental disparities. The approach is performed by a risk screening tool which employs an integrated multimedia exposure model able to manage all the steps of the analysis in a georeferenced framework operating in a GIS environment. In this study, we performed a risk assessment of metal exposure (cadmium, lead, selenium, arsenic, chromium, mercury and nickel) with a resolution of 1 km. The modeling was performed on a region of France (Nord-Pas de Calais) characterized by a heavy industrial past. The exposure pathways considered include ingestion of soil, vegetation, meat, egg, milk, fish and drinking water. Concentrations of metals on local food were calculated using a soil pollutants geodatabase and bioaccumulation food-chain models. We used water quality data and distance from river as a proxy for fish consumption exposure. Exposure scenarios were defined for different reference groups (age groups, anatomical and dietary properties, fraction of food produced locally). The results showed that the 2 to 7 age group is always the most vulnerable class of receptor. A spatial pattern analysis was performed to highlight hot-spot areas with significantly elevated risk values. In these areas, arsenic (up to 43 % among metals) and vegetables (up to 31 % among considered pathways) are the principal determinants of the calculated risk. In a next step, the same methodology will be use, adding inhalation pathway and organic pollutants, to map environmental inequalities and to analyse geographic cancer disparities in France.
Caracterisation topologique d'un jeu de donnes images avec les nombres de Betti et un modele generatif
Dans cet article, nous proposons un mod�ele g�en�eratif qui permet d'extraire les nombres de ... more Dans cet article, nous proposons un mod�ele g�en�eratif qui permet d'extraire les nombres de Betti d'un ensemble de vari�et�es de RD �a partir d'un �echantillon. Ce mod�ele est bas�e sur le Complexe Simplicial G�en�eratif, un mod�ele de m�elange dont les composantes sont des simplexes g�eom�etriques convolu�es �a une distribution gaussienne multivari�ee. La nouveaut�e de la m�ethode consiste �a optimiser BIC, un crit�ere statistique de vraisemblance p�enalis�ee, pour obtenir une estimation des invariants topologiques des vari�et�es g�en�eratrices des donn�ees, les nombres de Betti. Les r�esultats d'une telle m�ethode montrent une plus grande robustesse au bruit que ceux d'une m�ethode fond�ee sur une approche purement g�eom�etrique. Le Complexe Simplicial G�en�eratif est compar�e au Witness Complex (WitC) et la persistance homologique sur des donn�ees jouet (sph�ere, tore, bouteille de Klein) et un jeu de donn�ees r�eelles, COIL- 100, une base d'images d'...
A mixture of Kalman filters for online monitoring of railway switches
Assessing the operating state of the railway infrastructure and rolling stock using condition mea... more Assessing the operating state of the railway infrastructure and rolling stock using condition measurements acquired through embedded sensors has become a powerful decision-making support for preventive maintenance strategies. This article introduces a dynamic approach for the online monitoring of railway switch operations. The method is based on modeling the power consumption curves acquired during successive switch operations using conjointly five polynomial regression models whose coefficients are dynamically estimated across a sequence of curves. The experimental study conducted on two real power consumption curve sequences from the French high speed network has shown encouraging results in terms of characterization of the temporal evolution of railway switch operations.
Modèle de mélange poissonien pour l'analyse de données de déplacements
ABSTRACT Studies based on human mobility, including Bicycle Sharing System (BSS) trac analysis, h... more ABSTRACT Studies based on human mobility, including Bicycle Sharing System (BSS) trac analysis, has expanded over the past few years. They give insight of the underlying urban phenomena linked to city dynamics. This paper presents a generative count-series model using Poisson mixtures to automatically analyse and find temporal-based partitions over the Vélib' origin-destination (OD) flow-data. Such an approach may provide latent factors that reveal how regions of diffrent usage interact over the time.
Un graphe génératif pour la classification semi-supervisée
Ingénierie des systèmes d'information, 2010
We introduce a new semi-supervised algorithm based on a generative model. This model combines a G... more We introduce a new semi-supervised algorithm based on a generative model. This model combines a Gaussian mixture model and a generative graph built on the components of this mixture. The combination corresponds to refit the class membership of the mixture component with a propagation process. Both models can be optimized under the maximum likelihood framework and the only hyper-parameter (number of components of the mixture) can be selected with a statistical criterion. Experimental results show that we achieve accuracies comparable to those of rival state-of-the-art algorithms when few labeled data are available. Moreover, it offers the advantage of defining an objective statistical criterion for tuning its parameters, cancelling the need for arbitrary hand-tuning.
Uploads
Papers by Gérard Govaert