Mixture models

In recent years, video monitoring and surveillance systems have been widely used in traffic management. The image sequences for traffic scenes are recorded by a stationary camera. The video clip is sent to LabVIEW program to convert into image frames. NI LabVIEW vision assistant module is used to detect the moving vehicle. The method is based on the establishment of correspondences between regions and vehicles, as the vehicles move through the image sequence. Background subtraction is used which improves the adaptive background mixture model and makes the system learn faster and more accurately, as well as adapt effectively to changing environments. The resulting system robustly identifies vehicles, rejecting background and tracks vehicles over a specific period of time. Once the (object) vehicle is tracked, the attributes of the vehicle like width, length, perimeter, area etc are extracted by image process feature extraction techniques. In proposed system we use LabVIEW and Vision assistant module for image processing and feature extraction. The project will benefit to reduce cost of traffic monitoring system and complete automation of traffic monitoring system.

A suboptimal algorithm to fixed-interval smoothing for nonlinear Markovian switching systems is proposed. It infers a Gaussian mixture approximation to the posterior smoothing pdf by combining the statistics produced by an IMM filter into an original backward recursive process. The complexity is limited, as the number of underlying filters and smoothers is equal to the constant number of hypotheses in the posterior mixture. A comparison, conducted on realistic simulated target tracking case studies, shows that the investigated method performs significantly better than equivalent algorithms.

A suboptimal algorithm to fixed-interval smoothing for nonlinear Markovian switching systems is proposed. It infers a Gaussian mixture approximation to the posterior smoothing pdf by combining the statistics produced by an IMM filter into an original backward recursive process. The complexity is limited, as the number of underlying filters and smoothers is equal to the constant number of hypotheses in the posterior mixture. A comparison, conducted on realistic simulated target tracking case studies, shows that the investigated method performs significantly better than equivalent algorithms.

Whereas previous research has shown that either tree or spatial representations of dissimilarity judgments may be appropriate, focussing on the comparative fit at the aggregate level, we investigate whether there is heterogeneity among subjects in the extent to which their dissimilarity judgments are better represented by ultrametric tree or spatial multidimensional scaling models. We develop a mixture model for the analysis of dissimilarity data, that is formulated in a stochastic context, and entails a representation and a measurement model component. The latter involves distributional assumptions on the measurement error, and enables estimation by maximum likelihood. The representation component allows dissimilarity judgments to be represented either by a tree structure or by a spatial configuration, or a mixture of both. In order to investigate the appropriateness of tree versus spatial representations, the model is applied to twenty empirical data sets. We compare the fit of our model with that of aggregate tree and spatial models, as well as with mixtures of pure trees and mixtures of pure spaces, respectively. We formulate some empirical generalizations on the relative importance of tree versus spatial structures in representing dissimilarity judgments at the individual level.

Aphasia is the loss of the ability to produce and/or comprehend language, due to injury to brain areas responsible for these functions. Aphasic patients' performance on comprehension tests has traditionally been related both to the patient's individual ability and to the difficulty of the test questions. The natural choice for analysis of these test results is the Rasch model. It assumes that the probability of a patient responding correctly to a question is the inverse-logit function of the difference between the individual patient's ability and the difficulty of the test question. This study first modeled the way aphasic patients process different sentence types, as well as their ability to accomplish tasks using Rasch models. However, several scientifically important features of the data, such as the correlation of correct responses between two different comprehension tasks, and the association between response patterns in control sentences and response patterns in experimental sentences, were found to be inadequately captured by such models. Alternatively, we used a full Bayesian approach, exploring a mixture of generalized linear mixed models that clustered patients into similar response patterns and abilities. The mixture model was found to better describe the experimental results than any other model examined. The mixture model also expresses the hypothesis that aphasic patients can be classified into different ability and response profile groups, and that patients utilize different cognitive resources in different comprehension tasks. These results are scientifically important and could not have been discovered by using the simple Rasch model. This article has supplementary material online.

L'immunoprécipitation de la chromatine (ChIP) permet d'étudier les interactions entre les protéines et l'ADN ainsi que différents états chromatiniens. Le ChIP-chip est une technique combinant l'immunoprécipitation de la chromatine avec le principe des puces à ADN, ce qui permet une étude à l'échelle du génome. Nous nous intéressons ici à l'analyse des différences entre deux échantillons d'ADN immunoprécipité. Biologiquement, on s'attend à distinguer quatre groupes différents : un groupe d'ADN non-immunoprécipité, un groupe d'ADN immunoprécipité identiquement dans les deux échantillons et deux groupes dans lesquels l'ADN est immunoprécipité en quantités différentes. Nous modélisons ces données par un mélange de gaussiennes bidimensionnelles à quatre composants. Les matrices de variance sont contraintes afin d'intégrer des connaissances biologiques. Les paramètres sont estimés par l'algorithme EM. Nous appliquons cette méthode pour étudier la différence de méthylation d'une histone entre l'écotype sauvage de la plante modèle Arabidopsis thaliana et un mutant. Mots-clés : Mélange gaussien, décomposition spectrale, algorithme EM, ChIP-chip. Résumé Chromatin immunoprecipitation (ChIP) enables to investigate interactions between proteins and DNA and also various chromatin states. ChIP-chip is a well-established procedure combining chromatin immunoprecipitation with DNA microarrays, which allows a study of the whole genome. We are interested in the analyze of the differences between two immunoprecipitated DNA samples. From a biological point of view, we expect to distinguish four different groups : a group of non-immunoprecipited DNA, a group of immunoprecipited DNA in both samples, and then two groups in which DNA is differently immunoprecipited. We propose to model these data with a mixture of two-dimensional Gaussians with four components. Biological knowledges are included as constraints on the variance matrices. The parameters are estimated by the EM algorithm. This method is applied to Nim-bleGen data in order to study the histone methylation difference between the wild ecotype of the model plant Arabidopsis thaliana and a mutant.

Psychopathology, diagnosis, and classification of mental disorders have traditionally been based on a biomedical perspective. With the aim of defining and classifying mental disorders, the American Psychiatric Association (APA) and the World Health Organization (WHO) developed two systems that are widely used in the mental health field: the Diagnostic and Statistical Manual of Mental Disorders (DSM) and the International Classification of Diseases and Related Health Problems (ICD). However, the limitations of DSM and ICD have led to the development of alternative models, such as the Research Domain Criteria (RDoC), Network Theory (NT), the Power Threat Meaning Framework (PTMF), the Hierarchical Taxonomy of Psychopathology (HiTOP), and the Psychodynamic Diagnostic Manual (PDM- 2). This manuscript describes these models using posttraumatic stress disorder (PTSD) for exemplification, as it is one of the most challenging and controversial disorders, briefly expounding their strengths and limitations. Although none of the models have proposed a widely accepted conceptualization of posttraumatic stress, their combined use from an integrative approach could provide an accurate definition of this phenomenon. The aim of this paper is to disseminate these models among practitioners and academics and to foster a debate concerning the potential benefits of assuming an integrative approach instead of assuming models ascribed to specific theoretical frameworks that limit our understanding of psychiatric conditions.

1. Capture-recapture mixture models are important tools in evolution and ecology to estimate demographic parameters and abundance while accounting for individual heterogeneity. A key step is to select the correct number of mixture components i) to provide unbiased estimates that can be used as reliable proxies of fitness or ingredients in management strategies and ii) classify individuals into biologically meaningful classes. However, there is no consensus method in the statistical literature for selecting the number of components. 2. In ecology, most studies rely on the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC) that has recently gained attention in ecology. The Integrated Completed Likelihood criterion (ICL; IEEE Transactions on Pattern Analysis and Machine Intelligence, 2000, 22, 719) was specifically developed to favour well-separated components, but its use has never been investigated in ecology. 3. We compared the performance of AIC, BIC and ICL for selecting the number of components with regard to a) bias and accuracy of survival and detection estimates and b) success in selecting the true number of components using extensive simulations and data on wolf (Canis lupus) that were used for management through survival and abundance estimation. 4. Bias in survival and detection estimates was <0.02 for both AIC and BIC, and more than 0.09 for ICL, while mean square error was <0.05 for all criteria. As expected, bias increased as heterogeneity increased. Success rates of AIC and BIC in selecting the 'true' number of components were better than ICL (68% for AIC, 58% for BIC, and 16% for ICL). As the degree of heterogeneity increased, AIC (and BIC in a lesser extent) overestimated the number of components, while ICL often underestimated this number. For the wolf study, the 2-class model was selected by BIC and ICL, while AIC could not decide between the 2-and 3-class models. 5. We recommend using AIC or BIC when the aim is to estimate parameters. Regarding classification, we suggest taking the classification quality into account by using ICL in conjunction with BIC, pending further work to adapt its penalty term for capture-recapture data.

Managing large carnivores is one of the most controversial issues in wildlife conservation, as the sociopolitical challenges it raises are as important as the biological ones. Such controversial issues in wildlife conservation require objective biological components to be implemented within the management decision process, in particular, a reliable way of estimating trends in abundance. However, these species usually exhibit territoriality, low densities, and social constraints that can generate individual detection heterogeneity (IDH) of methodological (sampling) or biological (social status, marking behavior) origin. If not accounted for, IDH can lead, in turn, to strong bias in the estimation of population abundance. As a complement to population size, we propose to use the population growth rate (k) estimated with capture-recapture (CR) data, a robust method to detect and account for IDH, to monitor and manage elusive species. As a case study, we consider the natural recovery of the wolf (Canis lupus) population in France, for which a previous study has shown strong IDH leading to a 27% underestimation of abundance. Analyzing a nine-year data set based on a noninvasive sampling protocol relying on wolf scat genotyping, we adopt a new CR approach to estimate k while explicitly accounting for IDH. The annual population growth rate was estimated at 1.271 6 0.087 (mean 6 SE) vs. 1.270 6 0.095 when not accounting for IDH, indicating that k is much less sensitive to IDH than is abundance. We evaluated the robustness of our approach in a simulation study using increasing levels of IDH. The bias in k increased with detection heterogeneity but was low whether we used a model with IDH or not. Finally, we discuss the management implications of our findings in terms of sampling protocols and management strategies of elusive species.

Currently, diagnoses of psychopathology rely on discrete labels. The assumption is that each instance of a given disorder is importantly similar to other instances of that disorder and importantly distinct from instances of any other disorder. If two labels apply to one person, or two individuals with the same label show notable dissimilarities, that merely shows an instance of random variation or that the diagnostic criteria for the labels in question should be updated, while remaining discrete.

This assumption has come under attack from a variety of sources, leading to  proposals for new diagnostic frameworks, termed “transdiagnostic approaches” due to their emphasis on factors that cut across multiple  traditional diagnoses. I will focus on one leading example — the Hierarchical Taxonomy of Psychopathology (HiTOP) — showing how it plans to do away  with formal diagnostic labels in favor of a series of nested continua. While this proposal shows promise in improving the precision of diagnosis, it may come with a thorny social side effect, which has not been commented on in the literature.
To show this, I will give an overview of the notion of mental illness as a social identity. Specifically, I focus on affirmed communities founded on shared psychiatric diagnoses — that is, communities made up of members who actively and autonomously identify with their psychiatric condition. Notably, the boundaries of these communities and identities tend to follow the  boundaries of formal diagnostic labels. This system of discrete labels, then, however taxonomically unsound, provides the linguistic structure upon which marginalized social identities are built. Thus, it is worth asking: If we adopt a  transdiagnostic framework and get rid of discrete psychiatric labels, what happens to the communities and identities built atop those labels?

Because this matter is largely untouched in both the psychological and philosophical literatures, I will offer a tentative and preliminary framework for investigating this question. I will present and discuss three broad categories of potential impact. First, moving to a transdiagnostic framework may spell
doom for communities held together by psychiatric diagnoses. Without the backing of institutional medicine, potential members will fail to identify with the relevant psychiatric labels, and over time such identities will no longer be available in our social context. Second, these communities may live on with an increased barrier to entry, due to the increased obscurity. Recent  movements towards widened criteria for acceptance into these communities and the prevalence of self-diagnosis will be halted. I discuss ways in which this can be interpreted as beneficial or detrimental to those most in need of social support. Finally, it may lead to decreased gatekeeping across communities and renewed political consciousness. A rejection of inherent, implicitly biological, divisions between those with different diagnoses will open the door for coalition-building and a focus on social and political contributions to psychopathology.

Poverty measurement and the analysis of the progress (or otherwise) of the poor is beset with difficulties and controversies surrounding the definition of a poverty line or frontier. Here, using ideas from the partial identification literature and mixture models, a new approach to poverty measurement is proposed which avoids specifying a frontier, the price is that an agent's poverty status is only partially identified. Invoking variants of Gibrat's law to give structure to the distribution of outcomes for homogeneous subgroups of a population within the context of a finite mixture model of societal outcomes facilitates calculation of the probability of an agent's poverty status. From this it is straightforward to calculate all the usual poverty measures as well as other characteristics of the poor and non poor subgroups in a society. These ideas are exemplified in a study of 47 countries in Africa over the recent quarter century which reveals among other things a growing poverty rate and a growing disparity between poor and non poor groups not identified by conventional methods.

Poverty measurement and the analysis of the progress (or otherwise) of the poor is beset with difficulties and controversies surrounding the definition of a poverty line or frontier. Here, using ideas from the partial identification literature and mixture models, a new approach to poverty measurement is proposed which avoids specifying a frontier, the price is that an agent&#39;s poverty status

Covariance matrices of multivariate data capture feature correlations compactly, and being very robust to noise, they have been used extensively as feature descriptors in many areas in computer vision, like, people appearance tracking, DTI imaging, face recognition, etc. Since these matrices do not adhere to the Euclidean geometry, clustering algo-rithms using the traditional distance measures cannot be di-rectly extended to them. Prior work in this area has been re-stricted to using K-means type clustering over the Rieman-nian space using the Riemannian metric. As the applica-tions scale, it is not practical to assume the number of com-ponents in a clustering model, failing any soft-clustering algorithm. In this paper, a novel application of the Dirich-let Process Mixture Model framework is proposed towards unsupervised clustering of symmetric positive definite ma-trices. We approach the problem by extending the exist-ing K-means type clustering algorithms based on the logdet diver...

Phylogenetic analyses of DNA sequences were conducted to evaluate four alternative hypotheses of phrynosomatine sand lizard relationships. Sequences comprising 2871 aligned base pair positions representing the regions spanning ND1-COI and cyt b-tRNA Thr of the mitochondrial genome from all recognized sand lizard species were analyzed using unpartitioned parsimony and likelihood methods, likelihood methods with assumed partitions, Bayesian methods with assumed partitions, and Bayesian mixture models. The topology (Uma, (Callisaurus, (Cophosaurus, Holbrookia))) and thus monophyly of the ''earless" taxa, Cophosaurus and Holbrookia, is supported by all analyses. Previously proposed topologies in which Uma and Callisaurus are sister taxa and those in which Holbrookia is the sister group to all other sand lizard taxa are rejected using both parsimony and likelihood-based significance tests with the combined, unparitioned data set. Bayesian hypothesis tests also reject those topologies using six assumed partitioning strategies, and the two partitioning strategies presumably associated with the most powerful tests also reject a third previously proposed topology, in which Callisaurus and Cophosaurus are sister taxa. For both maximum likelihood and Bayesian methods with assumed partitions, those partitions defined by codon position and tRNA stem and nonstems explained the data better than other strategies examined. Bayes factor estimates comparing results of assumed partitions versus mixture models suggest that mixture models perform better than assumed partitions when the latter were not based on functional characteristics of the data, such as codon position and tRNA stem and nonstems. However, assumed partitions performed better than mixture models when functional differences were incorporated. We reiterate the importance of accounting for heterogeneous evolutionary processes in the analysis of complex data sets and emphasize the importance of implementing mixed model likelihood methods.

In Discrete Discriminant Analysis dimensionality problems often occur. In this context, we propose a combining models approach, taking profit from several potential models. In the bi-class case, a single combination coefficient is considered and estimated using several strategies. In the multi-class case, the decomposition into several bi-class problems embedded in a binary tree is implemented. New developments of this approach are presented and their performances assessed on real or simulated data.

We make an analogy between Man and the Grail because the former is the receptacle of energies of all kinds. Just as the Grail receives inexhaustible sources (water or light), so too the human being is a fine craftsman of the reception and then the distribution of these energies.

This paper introduces a novel method for feature selection in statistical models, combining the Minimum Average Variance Estimator (MAVE) with the Reciprocal Adaptive Bridge (RAB) penalty. The primary goal is to enhance variable selection in problems involving a large number of predictors while maintaining variance reduction. The proposed method leverages the benefits of Sufficient Dimension Reduction (SDR) along with the flexibility of the RAB penalty, allowing for dimension reduction with improved selectivity and accuracy in feature selection. Through a series of simulation studies and real-world application to medical data (Diabetes Dataset), the proposed method (SMAVE-RAB) is compared with other penalty methods such as Lasso and ALasso. Results from both simulation and application show that SMAVE-RAB outperforms the alternative methods in terms of Mean of Mean Squared Error (MMSE) and Standard Deviation (SD), highlighting its efficiency, accuracy, and stability when handling large datasets, particularly in high-dimensional medical settings.

Electromagnetic sensors such as ground penetrating radar and electromagnetic induction sensors are among the most widely used methods for the detection of buried land mines and unexploded ordnance. However, the performance of these sensors depends on the dielectric properties of the soil, which in turn are related to soil properties such as texture, bulk density, and water content. To predict the performance of electromagnetic sensors it is common to estimate the soil dielectric properties using models. However, the wide variety of available models, each with its own characteristics, makes it difficult to select the appropriate one for each occasion. In this paper we present an overview of the available methods, ranging from phenomenological Cole-Cole and Debye models to volume-based dielectric mixing models, and (semi-) empirical pedotransfer functions.

We introduce here the AdaptSgenoLasso, a new penalized likelihood method for gene mapping and for genomic prediction, which is an extended version of the SgenoLasso. The AdaptSgeno-Lasso relies on the original concept of a selective genotyping that varies along the genome. The "classical" selective genotyping on which the SgenoLasso is built on, consists in genotyping only extreme individuals, in order to increase the signal from genes. However, since the same amount of selection is applied at all genome locations, the signal is increased of the same proportional factor everywhere. With the AdaptSgenoLasso, we allow geneticists to impose more weights on some loci (i.e. locations) of interest, known to be responsible for the variation of the quantitative trait. The resulting signal is now dedicated to each locus. We propose here a deep theoretical study of the AdaptSgenoLasso, and we show on simulated data the superiority of this new approach over the SgenoLasso.

Laplace mixture model is widely used in lifetime applications. The estimation of model parameters is required to analyze the data. In this paper, the expectation maximization algorithm is used to obtain the estimates of parameters. The algorithm is a widely applicable approach to the iterative computation of the maximum likelihood estimates. However, even though the algorithm is useful for more than two components, we discuss it for a two components Laplace mixture model for simplicity. The behavior of the algorithm is explained with mathematical proofs. We also study the parameter estimates with respect to various sample sizes. The implementation of the expectation maximization algorithm is made via functions written in R script. The performance of the algorithm is guaranteed the convergent to a local maximum of the data log-likelihood model as a function of the model parameters. In addition, the results shown that the estimated parameters are closed to the real parameter values wh...

The limited availability of marine ingredients means that new and improved raw materials with high potential to replace fishmeal (FM) are required. Faba bean (Vicia faba) is a legume with good potential that has previously been tested in fish species with some promising results. The present study aimed to determine whether an air--classified faba bean protein concentrate (BPC, 615 g kg --1 crude protein content) could offer improved or favourable growth performance and physiological responses compared to the main commercially used protein sources, FM and soy protein concentrate (SPC), in Atlantic salmon (Salmo salar). The trial investigated the performance of 16 feeds formulated with varying FM/SPC/BPC proportions using a mixture design approach. Salmon parr of average weight 1.47 g were used as a model. The trial lasted eight weeks and also included high FM (560 g kg --1 ) and high defatted soybean meal (360 g kg --1 ) feeds as negative and positive controls respectively, for the assessment of gut inflammation. The results demonstrated conclusively that total inclusion levels of BPC ranging from 50 to 200 g kg --1 , partially replacing SPC and/or FM, displayed the greatest potential to be beneficial in terms of fish performance and nutrient composition with increased growth, protein content, fat content and ash. In addition to favourable whole--body composition parameters, it was found that inclusions of BPC below 340 g kg --1 of feed did not cause detrimental effects such as the gut inflammation observed in fish fed the high soybean meal control. High Inclusion level (450 g kg --1 ) of BPC caused a mild gut inflammation that was not as severe as that caused by the feed with high soybean meal. The results of this screening study indicate that BPC derived from faba beans can be a valuable alternative protein source in Atlantic salmon feeds. The data provided a platform to model the optimum range of BPC inclusion levels in combination with FM and SPC for further investigation in commercially relevant fish and conditions.

High-quality sources of protein for the formulation of feeds of carnivorous fish species such as Atlantic salmon are currently being sought. In an earlier screening trial we evaluated for the first time in Atlantic salmon (Salmo salar) the applicability of air-classified faba bean (Vicia faba) protein concentrate (BPC) inclusions in combination with soy protein concentrate (SPC) and fishmeal (FM) using parr as a model. Based on the results in parr in freshwater, the present study tested the hypothesis that BPC can effectively replace SPC and/or FM as a dietary protein source in post-smolt Atlantic salmon in seawater. The trial was designed to test nine different combinations of BPC, SPC and FM around the ideal mixture proportions modelled from data on parr (20% BPC, 46% SPC and 34% FM as percent of the total dietary protein). Somatic, physiological, biochemical and molecular parameters were assessed to determine fish performance. On average, fish grew from ~1.28 kg to 2.80 kg over the course of the 17 week trial. Results demonstrated that a mixture of BPC, SPC and FM (208 g kg -1 , 193 g kg -1 and 85 g kg -1 , respectively) resulted in the highest weight gain of post-smolts without significantly altering the FCR. Specifically, fish fed the highest inclusion of BPC and lowest level of FM had final weights on average 0.24 kg more than those with the lowest inclusion of BPC and highest FM. Improved weight gain was not achieved at the expense of poorer FCR, which was lower in fish fed diets with higher content of BPC and lower FM. The highest level of inclusion of BPC tested herein (~21% of feed) could replace up to 8.5% FM (from 17% to 8.5% of feed) and 10% SPC (from 29% to 19% of feed), resulting in the largest improvement in weight gain. There were no biologically significant or physiologically important effects on nutrient digestibility, enteritis scores or hepatic gene expression of dietary BPC. Statement of relevance: This study is highly relevant to the industry of feed manufacture.

The limited availability of marine ingredients means that new and improved raw materials with high potential to replace fishmeal (FM) are required. Faba bean (Vicia faba) is a legume with good potential that has previously been tested in fish species with some promising results. The present study aimed to determine whether an air--classified faba bean protein concentrate (BPC, 615 g kg --1 crude protein content) could offer improved or favourable growth performance and physiological responses compared to the main commercially used protein sources, FM and soy protein concentrate (SPC), in Atlantic salmon (Salmo salar). The trial investigated the performance of 16 feeds formulated with varying FM/SPC/BPC proportions using a mixture design approach. Salmon parr of average weight 1.47 g were used as a model. The trial lasted eight weeks and also included high FM (560 g kg --1 ) and high defatted soybean meal (360 g kg --1 ) feeds as negative and positive controls respectively, for the assessment of gut inflammation. The results demonstrated conclusively that total inclusion levels of BPC ranging from 50 to 200 g kg --1 , partially replacing SPC and/or FM, displayed the greatest potential to be beneficial in terms of fish performance and nutrient composition with increased growth, protein content, fat content and ash. In addition to favourable whole--body composition parameters, it was found that inclusions of BPC below 340 g kg --1 of feed did not cause detrimental effects such as the gut inflammation observed in fish fed the high soybean meal control. High Inclusion level (450 g kg --1 ) of BPC caused a mild gut inflammation that was not as severe as that caused by the feed with high soybean meal. The results of this screening study indicate that BPC derived from faba beans can be a valuable alternative protein source in Atlantic salmon feeds. The data provided a platform to model the optimum range of BPC inclusion levels in combination with FM and SPC for further investigation in commercially relevant fish and conditions.

In the wake of the novel coronavirus, SARS-CoV-19, the world has undergone a critical situation in which grave threats to global public health emerged. Among human populations across the planet, travel restraints, border enforcement measures, quarantine, and isolation provisions were implemented in an attempt to control and limit the spread of the contagion. Decisions on how to implement and enforce various control policies should be determined based on available real-world evidence and theoretical prediction. In this study, we propose a novel exportation- importation epidemic model associated with the quarantine and hospitalized-isolation policies by considering the two-body system: a source country of a contagious disease and a neighboring country that is initially disease-free. The model is first applied to the original COVID-19 data in China, Italy, and the Republic of Korea (ROK) and observed through consistent fitting results with equivalent goodness-of-fit. Then, the data are...

This paper considers the issue of modeling fractional data observed on [0,1), (0,1] or [0,1]. Mixed continuous-discrete distributions are proposed. The beta distribution is used to describe the continuous component of the model since its density can have quite different shapes depending on the values of the two parameters that index the distribution. Properties of the proposed distributions are examined. Also, estimation based on maximum likelihood and conditional moments is discussed. Finally, practical applications that employ real data are presented.

Recent research on machine learning focuses on audio source identification in complex environments. They rely on extracting features from audio signals and use machine learning techniques to model the sound classes. However, such techniques are often not optimized for a real-time implementation and in multi-source conditions. We propose a new real-time audio single-source classification method based on a dictionary of sound models (that can be extended to a multi-source setting). The sound spectrums are modeled with mixture models and form a dictionary. The classification is based on a comparison with all the elements of the dictionary by computing likelihoods and the best match is used as a result. We found that this technique outperforms classic methods within a temporal horizon of 0.5s per decision (achieved 6% of errors on a database composed of 50 classes). Future works will focus on the multi-sources classification and reduce the computational load.

Recent research on machine learning focuses on audio source identification in complex environments. They rely on extracting features from audio signals and use machine learning techniques to model the sound classes. However, such techniques are often not optimized for a real-time implementation and in multi-source conditions. We propose a new real-time audio single-source classification method based on a dictionary of sound models (that can be extended to a multi-source setting). The sound spectrums are modeled with mixture models and form a dictionary. The classification is based on a comparison with all the elements of the dictionary by computing likelihoods and the best match is used as a result. We found that this technique outperforms classic methods within a temporal horizon of 0.5s per decision (achieved 6% of errors on a database composed of 50 classes). Future works will focus on the multi-sources classification and reduce the computational load.

Computer network technology is developing quickly, and the advancement of internet techniques is growing faster. Furthermore, people and companies have became more aware of the importance of network security. To protect the network from different attacks, it is necessary to instantly detect intrusions with significantly low False-Positive Rates (FPRs). Many Anomalybased Detection Systems (ADS) have been proposed in the past. The performance of these systems depends on many factors such as features selection or extraction, missing or inaccurate records in data, over-fitting, under-fitting, high bias, and variance in data. Thus, it is important to take all these factors into account. Recently, a novel Geometric Area Analysis (GAA) technique based on Trapezoidal Area Estimation (TAE) has been proposed based on the Beta Mixture Model (BMM). In this work, we evaluate GAA and TAE techniques using other flexible mixture models based on inverted Beta, generalized Dirichlet, and generalized inverted Dirichlet distributions. The evaluation of this work is performed on two datasets, namely the NSL-KDD and UNSW-NB15. The results have shown the efficiency of the proposed ADS demonstrated by obtaining high accuracy and low false-positive rates in all attack types.

Impact experiments were performed at the french-german research Institute of Saint-Louis on three energetic materials composed of 70 % in weight of RDX particles embedded in a wax matrix. These materials differ by the microstructural properties of the explosive particles. The experimental results reveal that the detonation thresholds, and so the sensitivity to shock, are different for each sample. To better understand these results, we characterize the microstructural properties of these compositions. The microstructures of the three materials are imaged with micro-computed tomography (µCT) at CEA Gramat. First, the 3D µCT images are filtered and segmented. Then, an analysis of the segmented volume is performed (granulometry, covariance and weight fraction). Each labeled grain is extracted and characterized by several descriptors such as volume, surface, angularity, sphericity or intra-granular porosity. Virtual microstructures are generated with the extracted grains. Each grain is ...

In recent years, deep learning approaches have gained significant interest as a way of building hierarchical representations from unlabeled data. Additionally, in the field of sound direction-of-arrival (DOA) estimation, the binaural features like interaural time or phase difference and interaural level difference, or monaural cues like spectral peaks and notches are often used to estimate sound DOA. Although these binaural or monaural cues successfully explained human sound DOA ability, its accuracy and extent are all limited to the human knowledge on feature extracting methods. In this paper, we are interested in applying a deep learning approach to monaural sound localization based on monaural spectral cues. A convolutional deep belief network is applied to monaural auditory spectrograms processed by a computational auditory model to learning monaural features automatically. The learned features are then regressed using a support vector regression (SVR) model for sound DOA estimation tasks. Moreover, our feature representations learned from unlabeled monaural auditory spectrograms are then compared to the well-known binaural features. The results indicate that our monaural model shows reasonable performance at the task of 3D DOA estimation.

Wavelet-based demosaicing techniques have the advantage of being computationally relatively fast, while having a reconstruction performance that is similar to state-of-the-art techniques. Because the demosaicing rules are linear, it is fairly simple to integrate denoising into the demosaicing. In this paper, we present a method that performs joint denoising and demosaicing, using a Gaussian Scale Mixture (GSM) prior model, thereby modeling the local edge direction as a hidden variable. The results indicate that this technique offers a better reconstruction performance (in PSNR sense and visually) than sequential demosaicing and denoising. On a recent GPU, our algorithm takes 3.5 s for reconstructing a 12 megapixel RAW digital camera image.

To explore the "Perturb and Combine" idea for estimating probability densities, we study mixtures of tree structured Markov networks derived by bagging combined with the Chow and Liu maximum weight spanning tree algorithm and we try to accelerate the research procedure by reducing its computation complexity below the quadratic and keepingg similar accuracy. We empirically assess the performances of these heuristics in terms of accuracy and computation complexity, with respect to mixtures of bagged Markov trees, and single Markov tree CL built using the Chow and Liu algorithm.