LOGIN ENGLISH

제목 [학술세미나] [학과세미나] 10월 23일(월) 16시 학과세미나 안내
작성일 2017-10-18 08:52:28
내용 [세미나 안내]

▪제목 : Clustering aggregated symbolic data using pseudolikelihood
▪연사 : Junji Nakano (Professor, The Institute of Statistical Mathematics, JAPAN)
▪일시 : 2017년 10월 23일(월) PM 16:00 – 18:00
▪장소 : 25동 405호
[초 록] 
In recent “Big data” era, huge amount of individual data are available. They are often divided into some naturally defined groups, and sometimes we are mainly interested in the difference among groups, not among individuals. Such groups can be characterized and expressed by several descriptive statistics calculated by using information about the empirical joint distribution of variables. Symbolic data analysis is a method to handle such groups mainly by using the information of the marginal distribution. We propose to use up to second order moment statistics of each group and call them aggregated symbolic data. We consider that individual data is expressed by several categorical variables and continuous variables. By categorizing continuous variables and using pseudolikelihood, we define a dissimilarity measure between two groups, and use them for clustering all groups. Two real data examples are illustrated.


▪제목 : The A-hypergeometric distributions associated with the rational normal curve and some applications in Bayesian problems
▪연사 : Shuhei Mano (Professor, The Institute of Statistical Mathematics, JAPAN)
▪일시 : 2017년 10월 23일(월) PM 16:00 – 18:00
▪장소 : 25동 405호
[초 록] 
A distribution whose normalization constant is an A-hypergeometric polynomial is called an A-hypergeometric distribution. Such a distribution is in turn a generalization of the generalized hypergeometric distribution on the contingency tables with fixed marginal sums. In this talk, an exact sampling algorithm is presented for the general A-hypergeometric distributions, and the maximum likelihood estimation of the A-hypergeometric distribution associated with the rational normal curve is discussed in terms of the information geometry of Newton polytope. Then, two applications in Bayesian problems are discussed. The one is the conditional maximum likelihood estimation associated with Gibbs-type partitions. The other is sampling from measures on partitions. An exact sampler is an indispensable machinery in Bayesian mixture modeling. Urn schemes do not work without infinite exchangeability. Our exact sampler still works without infinite exchangeability.
 
파일 세미나 안내_171023(001_Junji Nakano)_A4.hwp [14KB]
세미나 안내_171023(002_Shuhei Mano)_A4.hwp [14KB]