重慶大學數(shù)學與統(tǒng)計學院邀請Wenyang Zhang教授作了一場題為“Model and Feature Selection in A Class of Semiparametric Models(型號和特征選擇的一類半參數(shù)模型)”的講座。數(shù)學與統(tǒng)計學院目前設有數(shù)學系、信息與計算科學系、統(tǒng)計與精算系,分別負責三個專業(yè)——數(shù)學與應用數(shù)學、信息與計算科學、統(tǒng)計學的建設與人才培養(yǎng)。講座的主要內容是:
選型是一首老歌的統(tǒng)計數(shù)據。隨著高維近年來的激增,人們開始用新調----的懲罰似然法播放。在這次演講中,我要研究一類半參數(shù)模型,其中潛在的解釋變量的數(shù)量增長比樣本量多快。我要提出一個新的處罰可能性程序,其選擇的重要特征,同時確定了正確的模式。我將探討懲罰部分建議過程中的有效性,并提出了一種新的方式把罰款。漸近性質將提交證明擬議的方法。我也表明了該程序的性能,當樣本量是有限的仿真研究。最后,我將用實際數(shù)據的例子說明了該方法的應用。
Wenyang Zhang教授主持的統(tǒng)計,數(shù)學系,約克英國的大學。副主編,雜志美國統(tǒng)計協(xié)會,2008年--會員中,IMS講義編委會 - 系列專著,2008-- 2010。會員,英國皇家統(tǒng)計學會的研究科委員會,2004-- 2008年。專業(yè)會員資格:_皇家統(tǒng)計學會會員。_美國統(tǒng)計協(xié)會會員。主要研究方向:高維數(shù)據分析;非參數(shù)建模;非線性時間序列;生存分析;功能性數(shù)據分析;多層次的模型;空間數(shù)據分析;結構方程模型。
原文:Model selection is an old song in statistics. With the surge of high dimensionality in recent years, people start to play it with a new tune ----the penalised likelihood method. In this talk, I am going to investigate a class of semiparametric models where the number of potential explanatory variables grows much fast than the sample size. I am going present a new penalised likelihood procedure which selects the important features and identifies the correct model simultaneously. I will explore the effectiveness of the penalty part in the proposed procedure, and present a new way to put penalty. Asymptotic properties will be presented to justify the proposed methodology. I will also show the performance of the proposed procedure when sample size is finite by simulation studies. Finally, I will illustrate the application of the proposed method by a real data example.