Bayesian nonparametric latent variable models

L’un des problèmes importants en apprentissage automatique est de déterminer la complexité du modèle à apprendre. Une trop grande complexité mène au surapprentissage, ce qui correspond à trouver des structures qui n’existent pas réellement dans les données, tandis qu’une trop faible complexité mène au sous-apprentissage, c’est-à-dire que l’expressivité du modèle est insuffisante pour capturer l’ensemble des structures présentes dans les données. Pour certains modèles probabilistes, la complexité du modèle se traduit par l’introduction d’une ou plusieurs variables cachées dont le rôle est d’expliquer le processus génératif des données. Il existe diverses approches permettant d’identifier le nombre approprié de variables cachées d’un modèle. Cette thèse s’intéresse aux méthodes Bayésiennes nonparamétriques permettant de déterminer le nombre de variables cachées à utiliser ainsi que leur dimensionnalité. La popularisation des statistiques Bayésiennes nonparamétriques au sein de la communauté de l’apprentissage automatique est assez récente. Leur principal attrait vient du fait qu’elles offrent des modèles hautement flexibles et dont la complexité s’ajuste proportionnellement à la quantité de données disponibles. Au cours des dernières années, la recherche sur les méthodes d’apprentissage Bayésiennes nonparamétriques a porté sur trois aspects principaux : la construction de nouveaux modèles, le développement d’algorithmes d’inférence et les applications. Cette thèse présente nos contributions à ces trois sujets de recherches dans le contexte d’apprentissage de modèles à variables cachées. Dans un premier temps, nous introduisons le Pitman-Yor process mixture of Gaussians, un modèle permettant l’apprentissage de mélanges infinis de Gaussiennes. Nous présentons aussi un algorithme d’inférence permettant de découvrir les composantes cachées du modèle que nous évaluons sur deux applications concrètes de robotique. Nos résultats démontrent que l’approche proposée surpasse en performance et en flexibilité les approches classiques d’apprentissage. Dans un deuxième temps, nous proposons l’extended cascading Indian buffet process, un modèle servant de distribution de probabilité a priori sur l’espace des graphes dirigés acycliques. Dans le contexte de réseaux Bayésien, ce prior permet d’identifier à la fois la présence de variables cachées et la structure du réseau parmi celles-ci. Un algorithme d’inférence Monte Carlo par chaîne de Markov est utilisé pour l’évaluation sur des problèmes d’identification de structures et d’estimation de densités. Dans un dernier temps, nous proposons le Indian chefs process, un modèle plus général que l’extended cascading Indian buffet process servant à l’apprentissage de graphes et d’ordres. L’avantage du nouveau modèle est qu’il admet les connections entres les variables observables et qu’il prend en compte l’ordre des variables. Nous présentons un algorithme d’inférence Monte Carlo par chaîne de Markov avec saut réversible permettant l’apprentissage conjoint de graphes et d’ordres. L’évaluation est faite sur des problèmes d’estimations de densité et de test d’indépendance. Ce modèle est le premier modèle Bayésien nonparamétrique permettant d’apprendre des réseaux Bayésiens disposant d’une structure complètement arbitraire.

Statistical learning of random probability measures

The study of random probability measures is a lively research topic that has attracted interest from different fields in recent years. In this thesis, we consider random probability measures in the context of Bayesian nonparametrics, where the law of a random probability measure is used as prior distribution, and in the context of distributional data analysis, where the goal is to perform inference given avsample from the law of a random probability measure. The contributions contained in this thesis can be subdivided according to three different topics: (i) the use of almost surely discrete repulsive random measures (i.e., whose support points are well separated) for Bayesian model-based clustering, (ii) the proposal of new laws for collections of random probability measures for Bayesian density estimation of partially exchangeable data subdivided into different groups, and (iii) the study of principal component analysis and regression models for probability distributions seen as elements of the 2-Wasserstein space. Specifically, for point (i) above we propose an efficient Markov chain Monte Carlo algorithm for posterior inference, which sidesteps the need of split-merge reversible jump moves typically associated with poor performance, we propose a model for clustering high-dimensional data by introducing a novel class of anisotropic determinantal point processes, and study the distributional properties of the repulsive measures, shedding light on important theoretical results which enable more principled prior elicitation and more efficient posterior simulation algorithms. For point (ii) above, we consider several models suitable for clustering homogeneous populations, inducing spatial dependence across groups of data, extracting the characteristic traits common to all the data-groups, and propose a novel vector autoregressive model to study of growth curves of Singaporean kids. Finally, for point (iii), we propose a novel class of projected statistical methods for distributional data analysis for measures on the real line and on the unit-circle.

Reasons For Brazilian Women To Switch From Different Contraceptives To Long-acting Reversible Contraceptives.

Long-acting reversible contraceptives (LARCs) include the copper-releasing intrauterine device (IUD), the levonorgestrel-releasing intrauterine system (LNG-IUS) and implants. Despite the high contraceptive efficacy of LARCs, their prevalence of use remains low in many countries. The objective of this study was to assess the main reasons for switching from contraceptive methods requiring daily or monthly compliance to LARC methods within a Brazilian cohort. Women of 18-50 years of age using different contraceptives and wishing to switch to a LARC method answered a questionnaire regarding their motivations for switching from their current contraceptive. Continuation rates were evaluated 1 year after method initiation. Sample size was calculated at 1040 women. Clinical performance was evaluated by life table analysis. The cutoff date for analysis was May 23, 2013. Overall, 1167 women were interviewed; however, after 1 year of use, the medical records of only 1154 women were available for review. The main personal reason for switching, as reported by the women, was fear of becoming pregnant while the main medical reasons were nausea and vomiting and unscheduled bleeding. No pregnancies occurred during LARC use, and the main reasons for discontinuation were expulsion (in the case of the IUD and LNG-IUS) and a decision to undergo surgical sterilization (in the case of the etonogestrel-releasing implant). Continuation rate was ~95.0/100 women/year for the three methods. Most women chose a LARC method for its safety and for practical reasons, and after 1 year of use, most women continued with the method.

Modeling of the Barkhausen jump in low carbon steel

This work presents a model for the magnetic Barkhausen jump in low carbon content steels. The outcomes of the model evidence that the Barkhausen jump height depends on the coercive field of the pinning site and on the mean free path of the domain wall between pinning sites. These results are used to deduce the influence of the microstructural features and of the magnetizing parameters on the amplitude and duration of the Barkhausen jumps. In particular, a theoretical expression, establishing the dependence of the Barkbausen jump height on the carbon content and grain size, is obtained. The model also reveals the dependence of the Barkhausen jump on the applied frequency and amplitude. Theoretical and experimental results are presented and compared, being in good agreement. (C) 2008 American Institute of Physics.

Galectin-1 Induces Reversible Phosphatidylserine Exposure at the Plasma Membrane

Cells normally undergo physiological turnover through the induction of apoptosis and phagocytic removal, partly through exposure of cell surface phosphatidylserine (PS). In contrast, neutrophils appear to possess apoptosis-independent mechanisms of removal. Here we show that Galectin-1 (Gal-1) induces PS exposure independent of alterations in mitochondrial potential, caspase activation, or cell death. Furthermore, Gal-1-induced PS exposure reverts after Gal-1 removal without altering cell viability. Gal-1-induced PS exposure is uniquely microdomain restricted, yet cells exposing PS do not display evident alterations in membrane morphology nor do they exhibit bleb formation, typically seen in apoptotic cells. Long-term exposure to Gal-1 prolongs PS exposure with no alteration in cell cycle progression or cell growth. These results demonstrate that Gal-1-induced PS exposure and subsequent phagocytic removal of living cells represents a new paradigm in cellular turnover.

Acute reversible inactivation of the bed nucleus of stria terminalis induces antidepressant-like effect in the rat forced swimming test

Background: The bed nucleus of stria terminalis (BNST) is a limbic forebrain structure involved in hypothalamo-pituitary-adrenal axis regulation and stress adaptation. Inappropriate adaptation to stress is thought to compromise the organism's coping mechanisms, which have been implicated in the neurobiology of depression. However, the studies aimed at investigating BNST involvement in depression pathophysiology have yielded contradictory results. Therefore, the objective of the present study was to investigate the effects of temporary acute inactivation of synaptic transmission in the BNST by local microinjection of cobalt chloride (CoCl(2)) in rats subjected to the forced swimming test (FST). Methods: Rats implanted with cannulae aimed at the BNST were submitted to 15 min of forced swimming (pretest). Twenty- four hours later immobility time was registered in a new 5 min forced swimming session (test). Independent groups of rats received bilateral microinjections of CoCl(2) (1 mM/100 nL) before or immediately after pretest or before the test session. Additional groups received the same treatment and were submitted to the open field test to control for unspecific effects on locomotor behavior. Results: CoCl(2) injection into the BNST before either the pretest or test sessions reduced immobility in the FST, suggesting an antidepressant-like effect. No significant effect of CoCl(2) was observed when it was injected into the BNST immediately after pretest. In addition, no effect of BNST inactivation was observed in the open field test. Conclusion: These results suggest that acute reversible inactivation of synaptic transmission in the BNST facilitates adaptation to stress and induces antidepressant-like effects.

Thunderclap headache attributed to reversible cerebral vasoconstriction: view and review

Thunderclap headache attributed to reversible cerebral vasoconstriction (THARCV) is a syndrome observed in a number of reported cases. In this article we reviewed this new headache entity (idiopathic form) using the clinical-radiological findings of 25 reported patients. In this series of patients 72% were women, the mean age at the onset of first headache episode was 39.4 +/- 2.3 years. In addition to the sine quanon condition of being abrupt and severe (thunderclap) at the onset, the headache was usually described as being explosive, excruciating, or crushing. The feature of pulsatility, accompanied or not by nausea was described by 80% of the patients. Forty percent of the cases manifested vomiting and 24% photophobia. Usually the headache was generalized, and in three cases it was unilateral at least at the onset. In 21 of 25 patients (84%) there was at least one recurrence or a sudden increase in the intensity of the headache. A past history of migraine was present in 52% of the patients. Precipitating factors were identified in 56% of the patients. Sexual intercourse was described by six patients. Of the 25 patients with THARCV syndrome studied, 12 (48%) developed focal neurological signs, transitory ischemic attack (n = 1), or ischemic stroke (n = 11, 44%), and two (8%) of them manifested seizures. The THARCV syndrome is a neurological disturbance perhaps more frequent than expected, preferentially affecting middle aged female migraineurs, and having an unpredictable prognosis, either showing a benign course or leading to stroke.

Positive Selection Results in Frequent Reversible Amino Acid Replacements in the G Protein Gene of Human Respiratory Syncytial Virus

Human respiratory syncytial virus (HRSV) is the major cause of lower respiratory tract infections in children under 5 years of age and the elderly, causing annual disease outbreaks during the fall and winter. Multiple lineages of the HRSVA and HRSVB serotypes co-circulate within a single outbreak and display a strongly temporal pattern of genetic variation, with a replacement of dominant genotypes occurring during consecutive years. In the present study we utilized phylogenetic methods to detect and map sites subject to adaptive evolution in the G protein of HRSVA and HRSVB. A total of 29 and 23 amino acid sites were found to be putatively positively selected in HRSVA and HRSVB, respectively. Several of these sites defined genotypes and lineages within genotypes in both groups, and correlated well with epitopes previously described in group A. Remarkably, 18 of these positively selected tended to revert in time to a previous codon state, producing a ""flipflop'' phylogenetic pattern. Such frequent evolutionary reversals in HRSV are indicative of a combination of frequent positive selection, reflecting the changing immune status of the human population, and a limited repertoire of functionally viable amino acids at specific amino acid sites.

The rating of perceived exertion predicts intermittent vertical jump demand and performance

The aims of this study were (a) to assess the ability of the rating of perceived exertion (RPE) to predict performance (i.e. number of vertical jumps performed to a fixed jump height) of an intermittent vertical jump exercise, and (b) to determine the ability of RPE to describe the physiological demand of such exercise. Eight healthy men performed intermittent vertical jumps with rest periods of 4, 5, and 6s until fatigue. Heart rate and RPE were recorded every five jumps throughout the sessions. The number of vertical jumps performed was also recorded. Random coefficient growth curve analysis identified relationships between the number of vertical jumps and both RPE and heart rate for which there were similar slopes. In addition, there were no differences between individual slopes and the mean slope for either RPE or heart rate. Moreover, RPE and number of jumps were highly correlated throughout all sessions (r=0.97-0.99; P0.001), as were RPE and heart rate (r=0.93-0.97; P0.001). The findings suggest that RPE can both predict the performance of intermittent vertical jump exercise and describe the physiological demands of such exercise.

Evaluation of an innovative critical power model in intermittent vertical jump

The aim of this study was to test if the critical power model can be used to determine the critical rest interval (CRI) between vertical jumps. Ten males performed intermittent countermovement jumps on a force platform with different resting periods (4.1 +/- 0.3 s, 5.0 +/- 0.4 s, 5.9 +/- 0.6 s). Jump trials were interrupted when participants could no longer maintain 95% of their maximal jump height. After interruption, number of jumps, total exercise duration and total external work were computed. Time to exhaustion (s) and total external work (J) were used to solve the equation Work = a + b . time. The CRI (corresponding to the shortest resting interval that allowed jump height to be maintained for a long time without fatigue) was determined dividing the average external work needed to jump at a fixed height (J) by b parameter (J/s). in the final session, participants jumped at their calculated CRI. A high coefficient of determination (0.995 +/- 0.007) and the CRI (7.5 +/- 1.6 s) were obtained. In addition, the longer the resting period, the greater the number of jumps (44 13, 71 28, 105 30, 169 53 jumps; p<0.0001), time to exhaustion (179 +/- 50, 351 +/- 120, 610 +/- 141, 1,282 +/- 417 s; p<0.0001) and total external work (28.0 +/- 8.3, 45.0 +/- 16.6, 67.6 +/- 17.8, 111.9 +/- 34.6 kJ; p<0.0001). Therefore, the critical power model may be an alternative approach to determine the CRI during intermittent vertical jumps.

Information Filtering and Array Algorithms for Discrete-Time Markovian Jump Linear Systems

This technical note develops information filter and array algorithms for a linear minimum mean square error estimator of discrete-time Markovian jump linear systems. A numerical example for a two-mode Markovian jump linear system, to show the advantage of using array algorithms to filter this class of systems, is provided.

Dependence of Barkhausen jump shape on microstructure in carbon steel

The present work presents measurements of the Magnetic Barkhausen Noise (MBN) in commercial AISI/SAE 1005 steel samples for different grain sizes. The correlation between the shape of the MBN jump and the grain size is established. The results show the existence of types of MBN jumps. Also, the outcome shows that one of these types of MBN jumps become ""squarer"" with the decrease of grain size.

A Separation Principle for the Continuous-Time LQ-Problem With Markovian Jump Parameters

In this paper, we devise a separation principle for the finite horizon quadratic optimal control problem of continuous-time Markovian jump linear systems driven by a Wiener process and with partial observations. We assume that the output variable and the jump parameters are available to the controller. It is desired to design a dynamic Markovian jump controller such that the closed loop system minimizes the quadratic functional cost of the system over a finite horizon period of time. As in the case with no jumps, we show that an optimal controller can be obtained from two coupled Riccati differential equations, one associated to the optimal control problem when the state variable is available, and the other one associated to the optimal filtering problem. This is a separation principle for the finite horizon quadratic optimal control problem for continuous-time Markovian jump linear systems. For the case in which the matrices are all time-invariant we analyze the asymptotic behavior of the solution of the derived interconnected Riccati differential equations to the solution of the associated set of coupled algebraic Riccati equations as well as the mean square stabilizing property of this limiting solution. When there is only one mode of operation our results coincide with the traditional ones for the LQG control of continuous-time linear systems.

Sampled Control for Mean-Variance Hedging in a Jump Diffusion Financial Market

In this technical note we consider the mean-variance hedging problem of a jump diffusion continuous state space financial model with the re-balancing strategies for the hedging portfolio taken at discrete times, a situation that more closely reflects real market conditions. A direct expression based on some change of measures, not depending on any recursions, is derived for the optimal hedging strategy as well as for the ""fair hedging price"" considering any given payoff. For the case of a European call option these expressions can be evaluated in a closed form.

On the Filtering Problem for Continuous-Time Markov Jump Linear Systems with no Observation of the Markov Chain

We consider in this paper the optimal stationary dynamic linear filtering problem for continuous-time linear systems subject to Markovian jumps in the parameters (LSMJP) and additive noise (Wiener process). It is assumed that only an output of the system is available and therefore the values of the jump parameter are not accessible. It is a well known fact that in this setting the optimal nonlinear filter is infinite dimensional, which makes the linear filtering a natural numerically, treatable choice. The goal is to design a dynamic linear filter such that the closed loop system is mean square stable and minimizes the stationary expected value of the mean square estimation error. It is shown that an explicit analytical solution to this optimal filtering problem is obtained from the stationary solution associated to a certain Riccati equation. It is also shown that the problem can be formulated using a linear matrix inequalities (LMI) approach, which can be extended to consider convex polytopic uncertainties on the parameters of the possible modes of operation of the system and on the transition rate matrix of the Markov process. As far as the authors are aware of this is the first time that this stationary filtering problem (exact and robust versions) for LSMJP with no knowledge of the Markov jump parameters is considered in the literature. Finally, we illustrate the results with an example.