題　目：Functional Prediction of Longitudinal Data

主講人：丘政民教授(中研院統計所)

時　間：100年6月10日(星期五)上午10:40-11:30
        (上午10:20-10:40茶會於交大統計所429室舉行)

地　點：交大綜合一館427室
--


Abstract

We discuss a functional prediction problem where one aims to predict yet
unobserved trajectories, given partial information available up to now. The
prediction is considered in the circumstances that one already observed a
sample of stochastic processes and the underlying random functions are
heterogeneously distributed. Concerning the heterogeneity characteristics of
the sample data, we consider probabilistic functional clustering based on
subspace projection coupled with a discriminative approach, assuming that the
random trajectories are sampled from a mixture of stochastic processes. A
functional linear regression model in conjunction with functional
discrimination is further developed for the prediction. The methods are
illustrated throughout an application to traffic flow analysis.

--
※ 編輯: wanting0605     來自: 140.113.252.176      (06/09 21:35)