Exponentiated gradient algorithms for large-margin structured classification

We consider the problem of structured classification, where the task is to predict a label y from an input x, and y has meaningful internal structure. Our framework includes supervised training of Markov random fields and weighted context-free grammars as special cases. We describe an algorithm that solves the large-margin optimization problem defined in [12], using an exponential-family (Gibbs distribution) representation of structured objects. The algorithm is efficient—even in cases where the number of labels y is exponential in size—provided that certain expectations under Gibbs distributions can be calculated efficiently. The method for structured labels relies on a more general result, specifically the application of exponentiated gradient updates [7, 8] to quadratic programs.

Comparison of Two Methods for Caluculating the Partition Functions of Various Spatial Statistical Models

Likelihood computation in spatial statistics requires accurate and efficient calculation of the normalizing constant (i.e. partition function) of the Gibbs distribution of the model. Two available methods to calculate the normalizing constant by Markov chain Monte Carlo methods are compared by simulation experiments for an Ising model, a Gaussian Markov field model and a pairwise interaction point field model.

Contexto e modularização em redes neurais recorrentes para aprendizagem de seqüências temporais

Este trabalho apresenta um sistema neural modular, que processa separadamente informações de contexto espacial e temporal, para a tarefa de reprodução de sequências temporais. Para o desenvolvimento do sistema neural foram considerados redes neurais recorrentes, modelos estocásticos, sistemas neurais modulares e processamento de informações de contexto. Em seguida, foram estudados três modelos com abordagens distintas para aprendizagem de seqüências temporais: uma rede neural parcialmente recorrente, um exemplo de sistema neural modular e um modelo estocástico utilizando a teoria de modelos markovianos escondidos. Com base nos estudos e modelos apresentados, esta pesquisa propõe um sistema formado por dois módulos sucessivos distintos. Uma rede de propagação direta (módulo estimador de contexto espacial) realiza o processamento de contexto espacial identificando a seqüência a ser reproduzida e fornecendo um protótipo do contexto para o segundo módulo. Este é formado por uma rede parcialmente recorrente (módulo de reprodução de sequências temporais) para aprender as informações de contexto temporal e reproduzir em suas saídas a seqüência identificada pelo módulo anterior. Para a finalidade mencionada, este mestrado utiliza a distribuição de Gibbs na saída do módulo para contexto espacial de forma que este forneça probabilidades de contexto espacial, indicando o grau de certeza do módulo e possibilitando a utilização de procedimentos especiais para os casos de dúvida. O sistema neural foi testado em conjuntos contendo trajetórias abertas, fechadas, e com diferentes situações de ambigüidade e complexidade. Duas situações distintas foram avaliadas: (a) capacidade do sistema em reproduzir trajetórias a partir de pontos iniciais treinados; e (b) capacidade de generalização do sistema reproduzindo trajetórias considerando pontos iniciais ou finais em situações não treinadas. A situação (b) é um problema de difícil ) solução em redes neurais devido à falta de contexto temporal, essencial na reprodução de seqüências. Foram realizados experimentos comparando o desempenho do sistema modular proposto com o de uma rede parcialmente recorrente operando sozinha e um sistema modular neural (TOTEM). Os resultados sugerem que o sistema proposto apresentou uma capacidade de generalização significamente melhor, sem que houvesse uma deterioração na capacidade de reproduzir seqüências treinadas. Esses resultados foram obtidos em sistema mais simples que o TOTEM.

A Test of Association between Qualitative Trait and a Set of SNPs

2000 Mathematics Subject Classification: 62P10, 92D10, 92D30, 62F03

Generalized KPCA by adaptive rules in feature space

Principal component analysis (PCA) is well recognized in dimensionality reduction, and kernel PCA (KPCA) has also been proposed in statistical data analysis. However, KPCA fails to detect the nonlinear structure of data well when outliers exist. To reduce this problem, this paper presents a novel algorithm, named iterative robust KPCA (IRKPCA). IRKPCA works well in dealing with outliers, and can be carried out in an iterative manner, which makes it suitable to process incremental input data. As in the traditional robust PCA (RPCA), a binary field is employed for characterizing the outlier process, and the optimization problem is formulated as maximizing marginal distribution of a Gibbs distribution. In this paper, this optimization problem is solved by stochastic gradient descent techniques. In IRKPCA, the outlier process is in a high-dimensional feature space, and therefore kernel trick is used. IRKPCA can be regarded as a kernelized version of RPCA and a robust form of kernel Hebbian algorithm. Experimental results on synthetic data demonstrate the effectiveness of IRKPCA. © 2010 Taylor & Francis.

Gibbs Energy of Formation of Cobalt Divanadium Tetroxide

The Gibbs energy of formation of V2O3-saturated spinel CoV2O4 has been measured in the temperature range 900–1700 K using a solid state galvanic cell, which can be represented as Pt, Co + CoV2O4 + V2O3/(CaO) ZrO2/Co + CoO, Pt. The standard free energy of formation of cobalt vanadite from component oxides can be represented as CoO (rs) + V2O3 (cor) → CoV2O4 (sp), ΔG° = −30,125 − 5.06T (± 150) J mole−1. Cation mixing on crystallographically nonequivalent sites of the spinel is responsible for the decrease in free energy with increasing temperature. A correlation between “second law” entropies of formation of cubic 2–3 spinels from component oxides with rock salt and corundum structures and cation distribution is presented. Based on the information obtained in this study and trends in the stability of aluminate and chromite spinels, it can be deduced that copper vanadite is unstable.

An improved method for calculating activities from distribution equilibria

The method of Gibbs-Duhem integration suggested by Speiser et al. has been modified to derive activities from distribution equilibria. It is shown that, in general, the activities of components in melts with a common anion can be calculated, without using their standard Gibbs energies of formation, from eqUilibrium ratios and the knowledge of activities in the metal phase. Moreover, if systems are so chosen that the concentration of one element in the metal phase lies in the Henry's law region (less than 1 %), information on activities in the metal phase is not required. Conversely, activities of elements in an alloy can be readily calculated from equilibrium distribution ratios alone, if the salt phase in equilibrium contains very small amounts of one element. Application of the method is illustrated using distribution ratios from the literature on AgCI-CuCI, AgBr-CuBr, and CuDo.5 -PbD systems. The results indicate that covalent bonding and van der Waals repulsive interactions in certain types of fused salt melts can significantly affect the thermodynamic properties of mixing.

Gibbs free energies of formation of ZnAl2O4 and ZnCr2O4

The standard free energies of formation of zinc aluminate and chromite were determined by measuring the oxygen potential over a solid CuZn alloy, containing 10 at.−% Zn, in equilibrium with ZnO, ZnAl2O4+Al2O3(χ) and ZnCr2O4+Cr2O3, in the temperature range 700–900°C. The oxygen potential was monitored by means of a solid oxide galvanic cell in which a Y2O3 ThO2 pellet was sandwiched between a CaOZrO2 crucible and tube. The temperature dependence of the free energies of formation of the interoxidic compounds can be represented by the equations, The heat of formation of the spinels calculated from the measurements by the “Second Law method” is found to be in good agreement with calorimetrically determined values. Using an empirical correlation for the entropy of formation of cubic spinel phases from oxides with rock-salt and corundum structures and the measured high temperature cation distribution in ZnAl2O4, the entropy of transformation of ZnO from wurtzite to rock-salt structure is evaluated.

Species transformation of trace elements and their distribution prediction in dyestuff residue incineration

The release of heavy metals from the combustion of hazardous wastes is an environmental issue of increasing concern. The species transformation characteristics of toxic heavy metals and their distribution are considered to be a complex problem of mechanism. The behavior of hazardous dyestuff residue is investigated in a tubular furnace under the general condition of hazardous waste pyrolysis and gasfication. Data interpretation has been aided by parallel theoretical study based on a thermodynamic equilibrium model based on the principle of Gibbs free energy minimization. The results show that Ni, Zn, Mn, and Cr are more enriched in dyestuff residue incineration than other heavy metals (Hg, As, and Se) subjected to volatilization. The thermodynamic model calculation is used for explaining the experiment data at 800 degrees C and analyzing species transformation of heavy metals. These results of species transformation are used to predict the distribution and emission characteristics of trace elements. Although most trace element predictions are validated by the measurements, cautions are in order due to the complexity of incineration systems.

Statistical thermodynamics of polydisperse polymer systems in the framework of lattice fluid model: Effect of molecular weight and its distribution on the spinodal in polymer solution

With the aid of thermodynamics of Gibbs, the expression of the spinodal was derived for the polydisperse polymer-solvent system in the framework of Sanchez-Lacombe Lattice Fluid Theory (SLLFT). For convenience, we considered that a model polydisperse polymer contains three sub-components. According to our calculation, the spinodal depends on both weight-average ((M) over bar (w)) and number-average ((M) over bar (n)) molecular weights of the polydisperse polymer, but the z-average molecular weight ((M) over bar (z)) dependence on the spinodal is invisible. The dependence of free volume on composition, temperature, molecular weight, and its distribution results in the effect of (M) over bar (n) on the spinodal. Moreover, it has been found that the effect of changing (M) over bar (w) on the spinodal is much bigger than that of changing (M) over bar (n) and the extrema of the spinodal increases with the rise of the weight-average molecular weight of the polymer in the solutions with upper critical solution temperature (UCST). However, the effect of polydispersity on the spinodal can be neglected for the polymer with a considerably high weight-average molecular weight. A more simple expression of the spinodal for the polydisperse polymer solution in the framework of SLLFT was also derived under the assumption of upsilon(*)=upsilon(1)(*)=upsilon(2)(*) and (1/r(1)(0))-(1/r(2i)(0))-->(1/r(1)(0)).

STATISTICAL THERMODYNAMICS IN THE FRAMEWORK OF THE LATTICE FLUID MODEL .1. POLYDISPERSE POLYMERS OF SPECIAL DISTRIBUTION

The Gibbs free energies and equations of state of polymers with special molar mass distributions, e.g., Flory distribution, uniform distribution and Schulz distribution, are derived based on a lattice fluid model. The influence of the polydispersity (or t

STATISTICAL THERMODYNAMICS IN THE FRAMEWORK OF THE LATTICE FLUID MODEL .2. BINARY MIXTURE OF POLYDISPERSE POLYMERS OF SPECIAL DISTRIBUTION

In this paper, the Gibbs free energy, the equation of state and the chemical potentials of polydisperse multicomponent polymer mixtures are derived. For general binary mixtures of polydisperse polymers, we also give the Gibbs free energy, the equation of

Bayesian Analysis for the Generalized Lognormal Distribution Applied to Failure Time Analysis

There are several versions of the lognormal distribution in the statistical literature, one is based in the exponential transformation of generalized normal distribution (GN). This paper presents the Bayesian analysis for the generalized lognormal distribution (logGN) considering independent non-informative Jeffreys distributions for the parameters as well as the procedure for implementing the Gibbs sampler to obtain the posterior distributions of parameters. The results are used to analyze failure time models with right-censored and uncensored data. The proposed method is illustrated using actual failure time data of computers.

Interação genótipo x ambiente para a produção de leite na espécie bubalina utilizando inferência Bayesiana por meio de Amostradores de Gibbs

Com o objetivo de verificar a existência da interação genótipo x ambiente, sob a forma de heterogeneidade de variâncias para a produção de leite na espécie bubalina e o seu impacto na avaliação genética dos animais, utilizando a inferência Bayesiana por meio de Amostrador de Gibbs, foram utilizados 5.484 registros de produção de leite referentes à produções de 2.994 búfalas predominantemente Murrah, filhas de 150 reprodutores, acasalados com 1130 matrizes, cujos partos ocorreram entre os anos de 1974 e 2004. Os registros foram provenientes do Programa de Melhoramento Genético dos Bubalinos (PROMEBUL) com a adição de registros provenientes do rebanho da EMBRAPA Amazônia Oriental -EAO, localizada em Belém, Pará. Foram estabelecidas classes de rebanho-ano de parto e de acordo com o desvio padrão de cada classe, os registros de produção de leite foram classificados em classes de alto e baixo desvio-padrão fenotípico. Posteriormente, os dados foram analisados desconsiderando e considerando as classes de desvio-padrão. O modelo utilizado empregou os efeitos fixos referentes às classes de rebanho-ano, mês de parto e covariáveis idade da fêmea ao parto e duração da lactação, além do efeito aleatório de animal, ambiente permanente e ambiente temporário. Para os efeitos fixos, foi assumido distribuição à priori uniforme e para os componentes de (co)variâncias foram assumidas distribuições priori qui-quadrado inversa e Wishart invertida. As médias observadas e desvio-padrão para produção de leite nas classes de alto e baixo desvio-padrão e em análise geral, foram iguais a 1870,21±758,78, 1900,50±587,76 e 1885,48±677,98, respectivamente. As médias posteriores para os componentes de variâncias foram maiores na classe de alto desvio-padrão. A herdabilidade obtida na classe de alto desvio-padrão foi próxima do valor observado na análise geral e inferior ao valor encontrado na classe de baixo desvio-padrão fenotípico. A correlação genética para produção de leite entre as classes de desvio-padrão foi igual a 0,58. As correlações de Spearman entre os valores genéticos para a produção de leite obtidos em análise geral com os valores obtidos nas classes de alto e baixo desvio padrão foram iguais a 0,94 e 0,93, respectivamente, para todos os reprodutores. Para uma amostra dos 10 melhores reprodutores, as mesmas correlações foram iguais a 0,94 e 0,47, respectivamente. Tais resultados revelam presença de heterogeneidade de variâncias entre rebanhos e esta heterogeneidade de variâncias é resultante de fatores ambientais, que podem levar a uma classificação errônea dos melhores reprodutores geneticamente para a produção leite.