報告題目：Fully Bayesian Inference for Structured Elastic Net

報告時間：2021年6月21日（周一）下午5:00

報告地點：9號樓513

報告人：王海斌教授

報告人單位：廈門大學(xué)

報告人簡介：

    王海斌，廈門大學(xué)數(shù)學(xué)科學(xué)學(xué)院教授、博士生導(dǎo)師。兼任中國現(xiàn)場統(tǒng)計研究會理事、中國現(xiàn)場統(tǒng)計研究會高維數(shù)據(jù)統(tǒng)計分會理事。主要從事潛在變量模型、非/半?yún)?shù)模型及時間序列分析的研究工作。曾主持國家自然科學(xué)基金面上項目和福建省自然科學(xué)基金面上項目、參與國家自然科學(xué)基金重點項目等。多次應(yīng)邀赴香港中文大學(xué)統(tǒng)計系進行合作研究。已在British Journal of Mathematical and Statistical Psychology、Computational Statistics and Data Analysis、Journal of Applied Probability、Journal of Time Series Analysis、Journal of Nonparametric Statistics、Psychometrika、Science China: Mathematics、Statistics and Computing等國內(nèi)外數(shù)學(xué)、概率、統(tǒng)計、心理學(xué)等主流學(xué)術(shù)期刊上發(fā)表學(xué)術(shù)論文30余篇。

報告摘要：

Structured elastic net is a rather general and flexible technique of regularization and variable selection, which includes the elastic net, the smooth lasso and the spline lasso as special cases. An appealing feature is that it can select groups of correlated predictors. We consider a fully Bayesian method to make statistical inference about it.  Main difficulty lies in that there exists an intractable term in the full conditional posterior of the tuning parameters, which makes ordinary MH algorithm unusable. We develop an exchange algorithm and a double MH sampler, respectively, to address this difficulty. We also consider an empirical posterior credible interval method with ``adaptively level'' for variable selection. The proposed methods are illustrated by the simulated examples, and applied to the diabetes and the biscuit dough datasets.

邀請單位：數(shù)學(xué)與統(tǒng)計學(xué)院