DOCSLIB.ORG

Student-T Process Regression with Student-T Likelihood

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Student-T Process Regression with Student-T Likelihood Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence (IJCAI-17) Student-t Process Regression with Student-t Likelihood Qingtao Tangy, Li Niuz, Yisen Wangy, Tao Daiy, Wangpeng Any, Jianfei Caiz, Shu-Tao Xiay y Department of Computer Science and Technology, Tsinghua University, China z School of Computer Science and Engineering, Nanyang Technological University, Singapore tqt15,dait14,wangys14,awp15 @mails.tsinghua.edu.cn; [email protected] f [email protected];g [email protected] Abstract distribution, GPR performs poorly on the data from heavy- tailed distributions or with outliers. However, real-world Gaussian Process Regression (GPR) is a powerful data often exhibit heavy-tailed phenomena [Nair et al., 2013] Bayesian method. However, the performance of and contain outliers [Bendre et al., 1994; Niu et al., 2015; GPR can be significantly degraded when the train- 2016]. ing data are contaminated by outliers, including In order to handle the outliers, heavy-tailed distribu- target outliers and input outliers. Although there tions (e.g., Laplace distribution, mixtures of Gaussians, and are some variants of GPR (e.g., GPR with Student-t Student-t distribution) have been introduced into GPR. In likelihood (GPRT)) aiming to handle outliers, most particular, Laplace noise is used in [Kuss, 2006] while mixed of the variants focus on handling the target outliers two forms of Gaussian corruption are used in [Naish-Guzman while little effort has been done to deal with the and Holden, 2008]. In [Neal, 1997; Vanhatalo et al., 2009; input outliers. In contrast, in this work, we aim to Jylanki¨ et al., 2011], the noise is assumed to follow the handle both the target outliers and the input outliers Student-t distribution (GPRT). However, all these methods at the same time. Specifically, we replace the Gaus- are only robust to the target outliers, but not robust to the sian noise in GPR with independent Student-t noise outliers in the inputs X, since the latent variables f(X) are to cope with the target outliers. Moreover, to en- still assumed to follow the Gaussian distribution. hance the robustness w.r.t. the input outliers, we use Related to the robustness w.r.t. the outliers in the inputs a Student-t Process prior instead of the common X, some works [Shah et al., 2014; Solin and Sarkk¨ a,¨ 2015; Gaussian Process prior, leading to Student-t Pro- Tang et al., 2016] rely on Student-t Process to handle the cess Regression with Student-t Likelihood (TPRT). input outliers. Particularly, the method in [Shah et al., 2014; We theoretically show that TPRT is more robust to Solin and Sarkk¨ a,¨ 2015] replaces the Gaussian Process with both input and target outliers than GPR and GPRT, the Student-t Process and incorporates the noise term into the and prove that both GPR and GPRT are special kernel function (TPRK) for computational simplicity. Fol- cases of TPRT. Various experiments demonstrate lowing [Shah et al., 2014; Solin and Sarkk¨ a,¨ 2015], an input that TPRT outperforms GPR and its variants on dependent Student-t noise (TPRD) is proposed in [Tang et both synthetic and real datasets. al., 2016]. Note that Tang et al. [2016] prove that TPRK, TPRD, and GPR have the same predictive mean if the kernel has a certain property named β property, which is actually 1 Introduction satisfied by most kernels. Taking the frequently used kernels Gaussian Process Regression (GPR) is a powerful Bayesian implemented by GPML [Rasmussen and Nickisch, 2010] (the method with good interpretability, non-parametric flexibil- most popular toolbox in Gaussian Process community) as ity, and simple hyper-parameter learning [Rasmussen, 2006]. examples, 24 out of 28 kernels have β property, for which Due to its nice properties, GPR has been successfully applied the above Student-t Process based methods (i.e., TPRK and to many fields, such as reinforcement learning [Rasmussen TPRD) have the same predictive value as GPR and thus fail et al., 2003], computer vision [Liu and Vasconcelos, 2015], to deal with the input outliers effectively. spatio-temporal modeling [Senanayake et al., 2016]. In this paper, with the aim to handle both the input outliers In GPR, the basic model is y = f(X) + , where y = and the target outliers at the same time, we propose Student- n n yi i=1 is the target vector, X = xi i=1 is the collection t Process Regression with Student-t Likelihood (TPRT). In off inputg vectors, and is the noise.f g The latent function our model, the latent function f is assumed to be a Student-t f is given a Gaussian Process prior and is assumed to Process prior while the noise is assumed to be an independent be independent and identically distributed (i.i.d.) Gaussian Student-t noise, instead of the noise incorporated into kernel noise. In practice, as the number of input vectors is finite, as in [Shah et al., 2014; Solin and Sarkk¨ a,¨ 2015] or dependent the latent variables f(X) follow a multivariate Gaussian noise as in [Tang et al., 2016]. In addition to owning all distribution. Due to the thin-tailed property of the Gaussian the advantages of GPR, such as good interpretability, non- 2822 Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence (IJCAI-17) parametric flexibility, and simple hyper-parameter learning, 2.2 Student-t Distribution and Student-t Process our proposed TPRT method is robust to both input and tar- get outliers, because the Student-t Process prior contributes The Student-t distribution [McNeil, 2006] we use in this to robustness to the input outliers while the independent paper is defined as follows. Definition 1. An n-dim random vector x Rn follows the Student-t noise assumption can cope with the target outliers. 2 One challenge of our TPRT method is that the inference is n-variate Student-t distribution with degrees of freedom ν n 2 R+, mean vector µ R , and correlation matrix R Π(n) analytically intractable. To solve the inference problem, we 2 2 utilize Laplace approximation for computing the posterior if its joint probability density function (PDF) is given by and marginal likelihood. The computational cost of TPRT Γ[(ν + n)=2] is roughly the same as that of GPRT, which also requires St (x ν; µ; R) = approximate inference. From the perspective of posterior and j Γ(ν=2)νn=2πn=2 R 1=2 j j marginal likelihood, we show that TPRT is more robust than − ν+n 1 2 GPR and GPRT. Besides, both GPR and GPRT are proved to 1 + (x µ)>R−1(x µ) : be special cases of TPRT. Finally, extensive experiments also · ν − − demonstrate the effectiveness of our TPRT method on both synthetic and real datasets. Given the definition of Student-t distribution, we can have the definition of Student-t Process [Shah et al., 2014]. 2 Background Definition 2. The process f is a Student-t Process (TP) on with degrees of freedom ν R+, mean function m: In this section, we will briefly introduce Gaussian Process RX, and kernel function k: 2 R if any finite subsetX! Regression (GPR), provide the definitions of Student-t dis- of function values have a multivariateX × X ! Student-t distribution, tribution and Student-t Process, and then compare Gaussian n i.e., f = f (xi) i=1 St(ν; µ; K) where K Π(n) with Process (GP) with Student-t Process (TP). f g ∼ n 2 Kij = k(xi; xj; θk) and µ R with µi = m(xi). We 2 2.1 Review of GPR denote that the process f is a Student-t Process with degrees of freedom ν, mean function m, and kernel function k as f In a regression problem, we have a training set = X; y ∼ n D f g TP (ν; m; k). of n instances, where X = xi and xi denotes a d-dim n f gi=1 input vector; y = yi i=1 and yi denotes a scalar output or target. In GPR, wef haveg 2.3 Comparison of GP and TP yi = f (xi) + i; i = 1; 2; :::; n; (1) In [Shah et al., 2014], it has been proved that GP is a special case of TP with degrees of freedom ν + . Among all the where i ; ; :::; n is assumed to be i.i.d. Gaussian i ( = 1 2 ) elliptical processes with an analytically! representable1 density, noise and the latent function f is given a GP prior, implying TP is the most general one, which implies its expressiveness that any finite subset of latent variables f f x n fol- = ( i) i=1 for nonparametric Bayesian modeling. The comparison of TP low a multivariate Gaussian distribution, i.e.,fp f Xg; K ( ) = and GP is illustrated in Figure 1 ([Shah et al., 2014]), from f µ; K , where µ is the mean vector andjK is the ( ) which we can see that TP allows the samples (blue solid) to covarianceN j matrix. Specifically, µ is an n-dim vector which be away from the mean (red dashed) while the samples of is usually assumed to be 0 for simplicity, and K is the GP gather around the mean. This indicates that the outliers covariance matrix with K k x ; x θ , in which k is i;j = ( i j; k) (usually away from the mean) will not have much effect on a kernel function and θ θ ; θ ; : : : ; θ is the set of k = ( k1 k2 kl) the mean of TP, but will affect the mean of GP severely as GP kernel parameters.
Load more

Recommended publications
  • Supplementary Materials
    Int. J. Environ. Res. Public Health 2018, 15, 2042 1 of 4 Supplementary Materials S1) Re-parametrization of log-Laplace distribution The three-parameter log-Laplace distribution, LL(δ ,,ab) of relative risk, μ , is given by the probability density function b−1 μ ,0<<μ δ 1 ab δ f ()μ = . δ ab+ a+1 δ , μ ≥ δ μ 1−τ τ Let b = and a = . Hence, the probability density function of LL( μ ,,τσ) can be re- σ σ τ written as b μ ,0<<=μ δ exp(μ ) 1 ab δ f ()μ = ya+ b δ a μδ≥= μ ,exp() μ 1 ab exp(b (log(μμ )−< )), log( μμ ) = μ ab+ exp(a (μμ−≥ log( ))), log( μμ ) 1 ab exp(−−b| log(μμ ) |), log( μμ ) < = μ ab+ exp(−−a|μμ log( ) |), log( μμ ) ≥ μμ− −−τ | log( )τ | μμ < exp (1 ) , log( ) τ 1(1)ττ− σ = μσ μμ− −≥τμμ| log( )τ | exp , log( )τ . σ S2) Bayesian quantile estimation Let Yi be the continuous response variable i and Xi be the corresponding vector of covariates with the first element equals one. At a given quantile levelτ ∈(0,1) , the τ th conditional quantile T i i = β τ QY(|X ) τ of y given X is then QYτ (|iiXX ) ii ()where τ iiis the th conditional quantile and β τ i ()is a vector of quantile coefficients. Contrary to the mean regression generally aiming to minimize the squared loss function, the quantile regression links to a special class of check or loss τ βˆ τ function. The th conditional quantile can be estimated by any solution, i (), such that ββˆ τ =−ρ T τ ρ =−<τ iiii() argmini τ (Y X ()) where τ ()zz ( Iz ( 0))is the quantile loss function given in (1) and I(.) is the indicator function.
    [Show full text]
  • Some Properties of the Log-Laplace Distribution* V
    SOME PROPERTIES OF THE LOG-LAPLACE DISTRIBUTION* V, R. R. Uppuluri Mathematics and Statistics Research Department Computer Sciences Division Union Carbide Corporation, Nuclear Division Oak Ridge, Tennessee 37830 B? accSpBficS of iMS artlclS* W8 tfcHteHef of Recipient acKnowta*g«9 th« U.S. Government's right to retain a non - excluslva, royalty - frw Jlcqns* In JMWl JS SOX 50j.yrt«ltt BfiXSflng IM -DISCLAIMER . This book w« prtMtM n in nonim nl mrk moniond by m vncy of At UnllM Sum WM»> ih« UWIrt SUM Oow.n™»i nor an, qprcy IMrw, w «v ol Ifmf mptovMi. mkn «iy w»ti«y, nptoi Of ImolM. « »u~i •ny U»l liability or tBpomlbllliy lof Ihe nary, ramoloweB. nr unhtlm at mi Womwlon, appmlui, product, or pram dliclowl. or rwromti that lu u« acuU r»i MMngt prhiltly ownad rljM<- flihmot Mi to any malic lomnwclal product. HOCOI. 0. »tvlc by ttada naw. tradtimr», mareittciunr. or otktmte, taa not rucaaainy coralltun or Imply In anoonamnt, nKommendailon. o» Iworlnj by a, Unilid SUM Oowrnment or m atncy limml. Th, Amnd ooMora ol «iihon ncrtwd fmtin dn not tauaaruv iu» or nlfcn thoaot tt» Unltad Sum GiMmrmt or any agmv trmot. •Research sponsored by the Applied Mathematical Sciences Research Program, Office of Energy Research, U.S. Department of Energy, under contract W-74Q5-eng-26 with the Union Carbide Corporation. DISTRIBii.il i yf T!it3 OGCUHEiU IS UHLIKITEOr SOME PROPERTIES OF THE LOS-LAPLACE DISTRIBUTION V. R. R. Uppuluri ABSTRACT A random variable Y is said to have the Laplace distribution or the double exponential distribution whenever its probability density function is given by X exp(-A|y|), where -» < y < ~ and X > 0.
    [Show full text]
  • Machine Learning - HT 2016 3
    Machine learning - HT 2016 3. Maximum Likelihood Varun Kanade University of Oxford January 27, 2016 Outline Probabilistic Framework � Formulate linear regression in the language of probability � Introduce the maximum likelihood estimate � Relation to least squares estimate Basics of Probability � Univariate and multivariate normal distribution � Laplace distribution � Likelihood, Entropy and its relation to learning 1 Univariate Gaussian (Normal) Distribution The univariate normal distribution is defined by the following density function (x µ)2 1 − 2 p(x) = e− 2σ2 X (µ,σ ) √2πσ ∼N Hereµ is the mean andσ 2 is the variance. 2 Sampling from a Gaussian distribution Sampling fromX (µ,σ 2) ∼N X µ By settingY= − , sample fromY (0, 1) σ ∼N Cumulative distribution function x 1 t2 Φ(x) = e− 2 dt √2π �−∞ 3 Covariance and Correlation For random variableX andY the covariance measures how the random variable change jointly. cov(X, Y)=E[(X E[X])(Y E[Y ])] − − Covariance depends on the scale of the random variable. The (Pearson) correlation coefficient normalizes the covariance to give a value between 1 and +1. − cov(X, Y) corr(X, Y)= , σX σY 2 2 2 2 whereσ X =E[(X E[X]) ] andσ Y =E[(Y E[Y]) ]. − − 4 Multivariate Gaussian Distribution Supposex is an-dimensional random vector. The covariance matrix consists of all pariwise covariances. var(X ) cov(X ,X ) cov(X ,X ) 1 1 2 ··· 1 n cov(X2,X1) var(X2) cov(X 2,Xn) T ··· cov(x) =E (x E[x])(x E[x]) = . . − − . � � . cov(Xn,X1) cov(Xn,X2) var(X n,Xn) ··· Ifµ=E[x] andΣ = cov[x], the multivariate normal is defined by the density 1 1 T 1 (µ,Σ) = exp (x µ) Σ− (x µ) N (2π)n/2 Σ 1/2 − 2 − − | | � � 5 Bivariate Gaussian Distribution 2 2 SupposeX 1 (µ 1,σ ) andX 2 (µ 2,σ ) ∼N 1 ∼N 2 What is the joint probability distributionp(x 1, x2)? 6 Suppose you are given three independent samples: x1 = 1, x2 = 2.7, x3 = 3.
    [Show full text]
  • Field Guide to Continuous Probability Distributions
    Field Guide to Continuous Probability Distributions Gavin E. Crooks v 1.0.0 2019 G. E. Crooks – Field Guide to Probability Distributions v 1.0.0 Copyright © 2010-2019 Gavin E. Crooks ISBN: 978-1-7339381-0-5 http://threeplusone.com/fieldguide Berkeley Institute for Theoretical Sciences (BITS) typeset on 2019-04-10 with XeTeX version 0.99999 fonts: Trump Mediaeval (text), Euler (math) 271828182845904 2 G. E. Crooks – Field Guide to Probability Distributions Preface: The search for GUD A common problem is that of describing the probability distribution of a single, continuous variable. A few distributions, such as the normal and exponential, were discovered in the 1800’s or earlier. But about a century ago the great statistician, Karl Pearson, realized that the known probabil- ity distributions were not sufficient to handle all of the phenomena then under investigation, and set out to create new distributions with useful properties. During the 20th century this process continued with abandon and a vast menagerie of distinct mathematical forms were discovered and invented, investigated, analyzed, rediscovered and renamed, all for the purpose of de- scribing the probability of some interesting variable. There are hundreds of named distributions and synonyms in current usage. The apparent diver- sity is unending and disorienting. Fortunately, the situation is less confused than it might at first appear. Most common, continuous, univariate, unimodal distributions can be orga- nized into a small number of distinct families, which are all special cases of a single Grand Unified Distribution. This compendium details these hun- dred or so simple distributions, their properties and their interrelations.
    [Show full text]
  • 4. Continuous Random Variables
    http://statwww.epfl.ch 4. Continuous Random Variables 4.1: Definition. Density and distribution functions. Examples: uniform, exponential, Laplace, gamma. Expectation, variance. Quantiles. 4.2: New random variables from old. 4.3: Normal distribution. Use of normal tables. Continuity correction. Normal approximation to binomial distribution. 4.4: Moment generating functions. 4.5: Mixture distributions. References: Ross (Chapter 4); Ben Arous notes (IV.1, IV.3–IV.6). Exercises: 79–88, 91–93, 107, 108, of Recueil d’exercices. Probabilite´ et Statistique I — Chapter 4 1 http://statwww.epfl.ch Petit Vocabulaire Probabiliste Mathematics English Fran¸cais P(A | B) probabilityof A given B la probabilit´ede A sachant B independence ind´ependance (mutually) independent events les ´ev´enements (mutuellement) ind´ependants pairwise independent events les ´ev´enements ind´ependants deux `adeux conditionally independent events les ´ev´enements conditionellement ind´ependants X,Y,... randomvariable unevariableal´eatoire I indicator random variable une variable indicatrice fX probability mass/density function fonction de masse/fonction de densit´e FX probability distribution function fonction de r´epartition E(X) expected value/expectation of X l’esp´erance de X E(Xr) rth moment of X ri`eme moment de X E(X | B) conditional expectation of X given B l’esp´erance conditionelle de X, sachant B var(X) varianceof X la variance de X MX (t) moment generating function of X, or la fonction g´en´eratrices des moments the Laplace transform of fX (x) ou la transform´ee de Laplace de fX (x) Probabilite´ et Statistique I — Chapter 4 2 http://statwww.epfl.ch 4.1 Continuous Random Variables Up to now we have supposed that the support of X is countable, so X is a discrete random variable.
    [Show full text]
  • Discriminating Between the Normal and the Laplace Distributions
    Discriminating Between The Normal and The Laplace Distributions Debasis Kundu1 Abstract Both normal and Laplace distributions can be used to analyze symmetric data. In this paper we consider the logarithm of the ratio of the maximized likelihoods to discriminate between the two distribution functions. We obtain the asymptotic distributions of the test statistics and it is observed that they are independent of the unknown parameters. If the underlying distribution is normal the asymptotic distribution works quite well even when the sample size is small. But if the underlying distribution is Laplace the asymptotic distribution does not work well for the small sample sizes. For the later case we propose a bias corrected asymptotic distribution and it works quite well even for small sample sizes. Based on the asymptotic distributions, minimum sample size needed to discriminate between the two distributing functions is obtained for a given probability of correct selection. Monte Carlo simulations are performed to examine how the asymptotic results work for small sizes and two data sets are analyzed for illustrative purposes. Key Words and Phrases: Asymptotic distributions; Likelihood ratio tests; Probability of correct selection; Location Scale Family. Address of Correspondence: 1 Department of Mathematics, Indian Institute of Tech- nology Kanpur, Pin 208016, INDIA. e-mail: [email protected]., Phone 91-512-2597141, Fax 91-512-2597500 1 1 Introduction Suppose an experimenter has n observations and the elementary data analysis, say for ex- ample histogram plot, stem and leaf plot or the box-plot, suggests that it comes from a symmetric distribution. The experimenter wants to ¯t either the normal distribution or the Laplace distribution, which one is preferable? It is well known that the normal distribution is used to analyze symmetric data with short tails, whereas the Laplace distribution is used for the long tails data.
    [Show full text]
  • Arxiv:2008.01277V3 [Q-Fin.RM] 13 Oct 2020
    GENERALIZED AUTOREGRESSIVE SCORE ASYMMETRIC LAPLACE DISTRIBUTION AND EXTREME DOWNWARD RISK PREDICTION APREPRINT Hong Shaopeng∗ School of Statistics Capital University of Economics and Trade BeiJing, China October 14, 2020 ABSTRACT Due to the skessed distribution, high peak and thick tail and asymmetry of financial return data, it is difficult to describe the traditional distribution. In recent years, generalized autoregressive score (GAS) has been used in many fields and achieved good results. In this paper, under the framework of generalized autoregressive score (GAS), the asymmetric Laplace distribution (ALD) is improved, and the GAS-ALD model is proposed, which has the characteristics of time-varying parameters, can describe the peak thick tail, biased and asymmetric distribution. The model is used to study the Shanghai index, Shenzhen index and SME board index. It is found that: 1) the distribution parameters and moments of the three indexes have obvious time-varying characteristics and aggregation characteristics. 2) Compared with the commonly used models for calculating VaR and ES, the GAS-ALD model has a high prediction effect. Keywords Generalized Autoregressive Score · Asymmetric Laplace Distribution · VaR · ES 1 Introduction As a special kind of economic system, the financial market has emerged with a series of unique macro laws, which are usually called "stylized facts". For example, the return distribution is sharp and asymmetrical. Due to Knight uncertainty, static parameter models usually cannot accurately describe the characteristics
    [Show full text]
  • Alpha-Skew Generalized T Distribution
    Revista Colombiana de Estadística July 2015, Volume 38, Issue 2, pp. 353 to 370 DOI: http://dx.doi.org/10.15446/rce.v38n2.51665 Alpha-Skew Generalized t Distribution Distribución t generalizada alfa sesgada Sukru Acitas1;a, Birdal Senoglu2;b, Olcay Arslan2;c 1Department of Statistics, Faculty of Science, Anadolu University, Eskisehir, Turkey 2Department of Statistics, Faculty of Science, Ankara University, Ankara, Turkey Abstract The alpha-skew normal (ASN) distribution has been proposed recently in the literature by using standard normal distribution and a skewing ap- proach. Although ASN distribution is able to model both skew and bimodal data, it is shortcoming when data has thinner or thicker tails than normal. Therefore, we propose an alpha-skew generalized t (ASGT) by using the gen- eralized t (GT) distribution and a new skewing procedure. From this point of view, ASGT can be seen as an alternative skew version of GT distribution. However, ASGT differs from the previous skew versions of GT distribution since it is able to model bimodal data sest as well as it nests most commonly used density functions. In this paper, moments and maximum likelihood estimation of the parameters of ASGT distribution are given. Skewness and kurtosis measures are derived based on the first four noncentral moments. The cumulative distribution function (cdf) of ASGT distribution is also ob- tained. In the application part of the study, two real life problems taken from the literature are modeled by using ASGT distribution. Key words: Bimodality, Kurtosis, Maximum Likelihood Estimation, Mod- eling, Skewness. Resumen La distribución normal alfa-sesgada (ASN por sus siglas en inglés) ha sido propuesta recientemente en la literatura mediante el uso de una distribución normal estándar y procedimientos de sesgo.
    [Show full text]
  • A Skew Laplace Distribution on Integers
    AISM (2006) 58: 555–571 DOI 10.1007/s10463-005-0029-1 Tomasz J. Kozubowski · Seidu Inusah A skew Laplace distribution on integers Received: 17 November 2004 / Revised: 24 February 2005 / Published online: 17 June 2006 © The Institute of Statistical Mathematics, Tokyo 2006 Abstract Wepropose a discrete version of the skew Laplace distribution. In contrast with the discrete normal distribution, here closed form expressions are available for the probability density function, the distribution function, the characteristic function, the mean, and the variance. We show that this distribution on integers shares many properties of the skew Laplace distribution on the real line, includ- ing unimodality, infinite divisibility, closure properties with respect to geometric compounding, and a maximum entropy property. We also discuss statistical issues of estimation under this model. Keywords Discrete Laplace distribution · Discrete normal distribution · Double exponential distribution · Exponential distribution · Geometric distribution · Geo- metric infinite divisibility · Infinite divisibility · Laplace distribution · Maximum entropy property · Maximum likelihood estimation 1 Introduction Any continuous distribution on R with p.d.f. f admits a discrete counterpart sup- ported on the set of integers Z ={0, ±1, ±2,...}. This discrete variable has p.m.f. of the form f(k) P(Y = k) = ∞ ,k∈ Z. (1) j=−∞ f(j) T.J. Kozubowski (B) Department of Mathematics and Statistics, University of Nevada, Reno, NV 89557, USA E-mail: [email protected] S. Inusah Department of Biostatistics, University of Alabama, Birmingham, AL 35294, USA 556 T.J. Kozubowski and S. Inusah Discrete normal (DN) distribution was mentioned in Lisman and van Zuylen (1972) in connection with maximum entropy property, and studied by Kemp (1997) (see also Dasgupta 1993; Szablowski 2001).
    [Show full text]
  • Functions by Name
    Title stata.com Functions by name abbrev(s,n) name s, abbreviated to a length of n abs(x) the absolute value of x acos(x) the radian value of the arccosine of x acosh(x) the inverse hyperbolic cosine of x age(ed DOB,ed ,snl ) the age in integer years on ed for date of birth ed DOB with snl the nonleap-year birthday for 29feb birthdates age frac(ed DOB,ed ,snl ) the age in years, including the fractional part, on ed for date of birth ed DOB with snl the nonleap-year birthday for 29feb birthdates asin(x) the radian value of the arcsine of x asinh(x) the inverse hyperbolic sine of x atan(x) the radian value of the arctangent of x atan2(y, x) the radian value of the arctangent of y=x, where the signs of the parameters y and x are used to determine the quadrant of the answer atanh(x) the inverse hyperbolic tangent of x autocode(x,n,x0,x1) partitions the interval from x0 to x1 into n equal-length intervals and returns the upper bound of the interval that contains x or the upper bound of the first or last interval if x < x0 or x > x1, respectively betaden(a,b,x) the probability density of the beta distribution, where a and b are the shape parameters; 0 if x < 0 or x > 1 binomial(n,k,θ) the probability of observing floor(k) or fewer successes in floor(n) trials when the probability of a success on one trial is θ; 0 if k < 0; or 1 if k > n binomialp(n,k,p) the probability of observing floor(k) successes in floor(n) trials when the probability of a success on one trial is p binomialtail(n,k,θ) the probability of observing floor(k) or more successes
    [Show full text]
  • Cauchy Principal Component Analysis
    Under review as a conference paper at ICLR 2015 CAUCHY PRINCIPAL COMPONENT ANALYSIS Pengtao Xie & Eric Xing School of Computer Science Carnegie Mellon University Pittsburgh, PA, 15213 fpengtaox,[email protected] ABSTRACT Principal Component Analysis (PCA) has wide applications in machine learning, text mining and computer vision. Classical PCA based on a Gaussian noise model is fragile to noise of large magnitude. Laplace noise assumption based PCA meth- ods cannot deal with dense noise effectively. In this paper, we propose Cauchy Principal Component Analysis (Cauchy PCA), a very simple yet effective PCA method which is robust to various types of noise. We utilize Cauchy distribution to model noise and derive Cauchy PCA under the maximum likelihood estimation (MLE) framework with low rank constraint. Our method can robustly estimate the low rank matrix regardless of whether noise is large or small, dense or sparse. We analyze the robustness of Cauchy PCA from a robust statistics view and present an efficient singular value projection optimization method. Experimental results on both simulated data and real applications demonstrate the robustness of Cauchy PCA to various noise patterns. 1 INTRODUCTION Principal Component Analysis and related subspace learning techniques based on matrix factoriza- tion have been widely used in dimensionality reduction, data compression, image processing, feature extraction and data visualization. It is well known that PCA based on a Gaussian noise model (i.e., Gaussian PCA) is sensitive to noise of large magnitude, because the effects of large noise are ex- aggerated by the use of Gaussian distribution induced quadratic loss. To robustify PCA, a number of improvements have been proposed (De La Torre & Black, 2003; Khan & Dellaert, 2004; Ke & Kanade, 2005; Archambeau et al., 2006; Ding et al., 2006; Brubaker, 2009; Candes et al., 2009; Eriksson & Van Den Hengel, 2010).
    [Show full text]
  • Alpha-Skew Generalized Normal Distribution and Its Applications
    Available at http://pvamu.edu/aam Applications and Applied Appl. Appl. Math. Mathematics: ISSN: 1932-9466 An International Journal (AAM) Vol. 14, Issue 2 (December 2019), pp. 784 - 804 Alpha-Skew Generalized Normal Distribution and its Applications 1*Eisa Mahmoudi, 2Hamideh Jafari and 3RahmatSadat Meshkat Department of Statistics Yazd University Yazd, Iran [email protected]; [email protected]; [email protected] * Corresponding Author Received: July 8, 2018; Accepted: November 11, 2019 Abstract The main object of this paper is to introduce a new family of distributions, which is quite flexible to fit both unimodal and bimodal shapes. This new family is entitled alpha-skew generalized normal (ASGN), that skews the symmetric distributions, especially generalized normal distribution through this paper. Here, some properties of this new distribution including cumulative distribution function, survival function, hazard rate function and moments are derived. To estimate the model parameters, the maximum likelihood estimators and the asymptotic distribution of the estimators are discussed. The observed information matrix is derived. Finally, the flexibility of the new distribution, as well as its effectiveness in comparison with other distributions, are shown via an application. Keywords: Alpha-skew distribution; Generalized normal distribution; Maximum likelihood estimator; Skew-normal distribution MSC 2010 No.: 60E05, 62F10 1. Introduction Recently, the skew-symmetric distributions have received considerable amount of attention, for example, the alpha skew normal (ASN) distribution is studied by Elal-Olivero (2010). First, he introduced a new symmetric bimodal-normal distribution with the pdf 784 AAM: Intern. J., Vol. 14, Issue 2 (December 2019) 785 f( y ) y2 ( y ), where is standard normal density function, and then defined alpha-skew-normal distribution as (1 y )2 1 f( y ; ) ( y ), y ¡¡, , 2 2 where is skew parameter.
    [Show full text]