Minimum error entropy (MEE) is an information theoretic learning approach that minimizes the contained in prediction error, which measured by entropy. It has been successfully used various machine tasks for its robustness to heavy-tailed distributions and outliers. In this paper, we consider use nonparametric regression analyze generalization performance from a theory perspective imposing [Formula: see text]th order moment condition on noise variable. To end, establish comparison theorem...

In a regression setup, we study in this brief the performance of Gaussian empirical gain maximization (EGM), which includes broad variety well-established robust estimation approaches. particular, conduct refined learning theory analysis for EGM, investigate its calibration properties, and develop improved convergence rates presence heavy-tailed noise. To achieve these purposes, first introduce new weak moment condition that could accommodate cases where noise distribution may be...

This paper is concerned with the testing hypotheses of regression parameters in linear models which errors are negatively superadditive dependent (NSD). A robust M-test base on M-criterion proposed. The asymptotic distribution test statistic obtained and consistent estimates redundancy involved established. Finally, some Monte Carlo simulations given to substantiate stability parameter power test, for various choices M-methods, explanatory variables different sample sizes.

In this paper, we provide some new continued fraction approximation and inequalities of the Somos quadratic recurrence constant, using its relation with generalized Euler constant.

The methods commonly used to test the associations between ordinal phenotypes and genotypes often treat either phenotype or genotype as continuous variables. To address limitations of these approaches, we propose a model where both are viewed manifestations an underlying multivariate normal random variable. proposed method allows modeling phenotype, covariates jointly. We employ generalized estimating equation technique M-estimation theory estimate parameters deduce corresponding asymptotic...

The problem of ranking/ordering instances, instead simply classifying them, has recently gained much attention in machine learning. Ranking from binary comparisons is a ubiquitous modern learning applications. In this paper, we consider ℓ 1 -norm SVM for ranking. As well known, with restrictions usually leads to sparsity. Moreover, independently draw sample sequence, are given exponentially strongly mixing sequence. Under some mild conditions, rate established.

Deterministic differential Tomographic SAR (D-TomoSAR) model, based on geometrical derivations and the assumption of accurate phase calibration, is widely employed for spatially locating temporally monitoring point-like scatterers in past.In this work, we model miscalibration effects extended scatters caused by partial correlation, i.e., decorrelation from temporal spatial changes as well residual atmospheric deformation effect after preprocessing.Starting origin 4-D focusing, correlation...

Let a = ( 1 , 2 …, m )∈ ℂ be an ‐dimensional vector. Then, it can identified with × circulant matrix. By using the theory of matrix‐valued wavelet analysis (Walden and Serroukh, 2002), we discuss vector‐valued multiresolution analysis. Also, derive several different designs finite length filters. The corresponding scaling functions are given. Specially, deal construction filters on symmetric space.