活動取消」專題演講 主講人:洪弘教授(台灣大學流行病學與預防醫學研究所)

  • 2021-05-14
  • AdminAdmin
  • 演講者:  /  主持人:
This is an image


因應疫情升溫,本次(1100514)學術演講活動取消。

題 目:A generalized information criterion for high-dimensional PCA rank selection

主講人:洪弘教授(台灣大學流行病學與預防醫學研究所)

時 間:110年5月14日(星期五)上午10:40-11:30

(上午10:20-10:40茶會於交大統計所428室舉行) 

地 點:交大綜合一館427室
 
摘要
 
Principal component analysis (PCA) is a commonly used statistical procedure for dimension reduction. An important issue for PCA is to determine the rank, which is the number of dominant eigenvalues of the covariance matrix. Among information-based criteria, Akaike information criterion (AIC) and Bayesian information criterion (BIC) are two most common ones. Both use the number of free parameters for assessing model complexity, which may suffer the problem of model misspecification. To alleviate this difficulty, we propose using the generalized information criterion (GIC) for PCA rank selection. The resulting GIC model complexity takes into account the sizes of eigenvalues and, hence, is more robust to model misspecification. The asymptotic properties and selection consistency of GIC are derived under the high-dimensional setting. Compared to AIC and BIC, the proposed GIC is better capable than AIC in excluding noise eigenvalues, and is more sensitive than BIC in detecting signal eigenvalues. Moreover, we discuss an application of GIC to selecting the number of factors for factor analysis. Our numerical study reveals that GIC compares favorably to the methods based on (deterministic) parallel analysis. 
 


使用Google Meet線上直播,以講者簡報畫面加聲音方式
演講開始前20分鐘可進入會議,請點選下列連結後按下「要求加入」即可