主题:High-Dimensional Online Model Averaging for Streaming Data
主讲人:王江峰 浙江工商大学
主持人:王国长 暨南大学
时间:2026年5月7日(周四)下午16:00-17:00
地点:暨南大学石牌校区经济学院大楼(中惠楼)102室
摘要
We propose a distribution-free online model averaging method for high-dimensional streaming data. Based on renewable least-squares summary statistics, the proposed method recursively updates the candidate-model estimators and constructs an online AIC or BIC-type criterion for weight selection. To ensure computational feasibility in high-dimensional streaming settings, we build a sparse nested candidate model family through marginal-correlation ranking and sparse-grid accumulation. When all candidate models are misspecified, the proposed method is asymptotically optimal in terms of relative squared loss. When the candidate family contains at least one unbiased model, the candidate-model estimators and the variance estimator are consistent, and the estimated weights asymptotically concentrate on the unbiased candidate models, with the BIC-type criterion further degenerating to the smallest unbiased model. Simulation studies show that the proposed method is more stable than the baseline online linear model, while the sparse-grid implementation delivers substantial computational savings with little loss of statistical accuracy. A semi-synthetic analysis based on the superconductivity dataset further demonstrates the practical effectiveness of the proposed method.
主讲人简介

王江峰,博士(后),浙江工商大学统计学教授,博士生导师,校西湖学者拔尖人才,应用概率统计研究所所长。全国工业统计学教学研究会理事,中国优选法统筹法与经济数学研究会数据科学分会理事。2010年同济大学数理统计专业毕业,获博士学位;2012年同济大学经管学院博士后出站;曾在加拿大滑铁卢大学统计与精算系访问一年,在华东师范大学统计学院访问一年。目前研究方向:时空数据分析、复杂生存数据分析、高维数据分析、分位数回归方法、经验似然方法等。主持国家社科基金一般项目3项、教育部人文社科基金等省部级科研项目5项,并主持省级教改项目2项;在《中国科学》《数学学报》(中英文版)、《J. R. Stat. Soc. Ser. C-Appl. Stat》《IEEE Trans. Cybern》《Statist. Computing》《J. Statist. Plann. Inference》《Statist. Papers》等发表论文50余篇。
校对 |王国长
责编 | 彭毅
初审 | 姜云卢
终审发布 | 何凌云
(来源:暨南大学经济学院微信公众号)

