清华大学、交通大学 统 计 学 研 究 所 专 题 演 讲 题 目: Model Selection Using Generalized Degrees of Freedom 主讲人: 黄信诚博士 (中央研究院 统计科学研究所) 时 间: 96年5月11日（星期五）10：40 - 11：30 （上午10：20-10：40茶会於统计所821室举行） 地 点: 清大综合三馆837室 Abstract Model selection is important in many scientific and engineering problems. In the literature, model selection in the context of nonGaussian distributions or among nonlinear procedures has not yet received a lot of attention. In this talk, a general technique of model assessment based on generalized degrees of freedom (GDF) is introduced and a formula of unbiased risk estimation is derived, which leads to a methodology of estimating GDF via data perturbation. The methodology is allowed to compare arbitrary complex methods regardless of whether the candidate methods are parametric or nonparametric, and whether the estimates are linear or nonlinear. Some numerical examples on geostatistical models and Poisson regression models will be provided, and some theoretical justification will be given to demonstrate the effectiveness of the proposed methodology. --

※ 发信站: 批踢踢实业坊(ptt.cc) ◆ From: 140.113.114.178