時間：2022年5月13日，星期五，上午

報告一(9:00---9:45)

題目：Multiple Structural Break Estimations for Linear Regression with Dependent Observations

摘要：Linear model with multiple change-point appears in a vast amount of statistical and econometric applications. In this article, the change-point detection problem is transformed into a variable selection problem by segmenting the data series and establishing a high-dimensional linear regression model with dependent observations. We apply Group Orthogonal Greedy Algorithm (GOGA) in variable selection to adapt to the situation that the number of change-point increases with the number of observations, adding high-dimensional information criteria (HDIC) to prevent overfitting. In the first step, GOGA+HDIC+Trim is applied to the segmented data for variable selection, significantly reducing the calculation cost. The quasi-likelihood ratio test in the second step can obtain a more accurate changepoint position. Under mild conditions, we prove the consistency of the number and the location of change-point. The simulation results and real data application also demonstrate the effectiveness of the algorithm.

報告人簡介：金百鎖，中國科學技術大學管理學院統(tǒng)計與金融系教授。2001年畢業(yè)于中國科學技術大學獲得學士學位，2006年獲得中國科學技術大學博士學位。研究方向為變結構模型，隨機矩陣，空間統(tǒng)計等。在PNAS，AoS，Biometrika，AAP等期刊已發(fā)表學術論文近50篇。先后主持國家自然科學基金青年項目、面上項目、國際交流項目和安徽省自然科學基金杰青項目?，F(xiàn)為中國現(xiàn)場統(tǒng)計研究會理事，全國工業(yè)統(tǒng)計學教學研究會理事，中國現(xiàn)場統(tǒng)計研究會資源與環(huán)境統(tǒng)計分會理事，中國現(xiàn)場統(tǒng)計研究會旅游大數(shù)據(jù)分會常務理事、副理事長。個人主頁：http://bs.ustc.edu.cn/Chinese/Profile-91.html

報告二 (9:50---10:35)

題目：Optimal Parameter-Transfer Learning by Model Averaging under Semiparametric Models

摘要：Transfer learning has attracted more and more attention in the field of artificial intelligence, of which the aim is to improve one target task of interest by utilizing tasks from several related source domains. In this article, we focus on the prediction for semiparametric additive linear model under the setting of transfer learning. Inheriting the spirits of parameter-transfer learning, we assume existing common knowledge shared in parametric components among different models in our framework that is possibly helpful for the target predictive task. We adopt a frequentist model averaging strategy to utilize parameter information. The theoretical properties have also been established, including the asymptotic optimality based on out-of-sample prediction risk and the property of weight convergence under some regularity conditions. Extensive numerical results demonstrate the superiority of the proposed method under various simulation designs comparing with competitive methods. (Jointly with Xiaonan Hu).

報告人簡介：張新雨，中科院數(shù)學與系統(tǒng)科學研究院預測中心研究員，中科大管理學院雙聘教授。主要從事計量經(jīng)濟學和統(tǒng)計學的理論和應用研究工作，具體研究方向包括模型平均、機器學習和組合預測等。擔任期刊《JSSC》領域主編、期刊《系統(tǒng)科學與數(shù)學》、《數(shù)理統(tǒng)計與管理》等的編委，是雙法學會數(shù)據(jù)科學分會副理事長、國際統(tǒng)計學會當選會員，先后主持自科優(yōu)秀和杰出青年基金項目。個人主頁：https://bs.ustc.edu.cn/Chinese/profile-578.html

報告三 (10:40---11:25)

題目：Statistically Guided Divide-and-Conquer for Sparse Factorization of Large Matrix

摘要：The sparse factorization of a large matrix is fundamental in modern statistical learning. In particular, the sparse singular value decomposition and its variants have been utilized in multivariate regression, factor analysis, biclustering, vector time series modeling, among others. The appeal of this factorization is owing to its power in discovering a highly-interpretable latent association network, either between samples and variables or between responses and predictors. However, many existing methods are either ad hoc without a general performance guarantee, or are computationally intensive, rendering them unsuitable for large-scale studies. We formulate the statistical problem as a sparse factor regression and tackle it with a divide-and-conquer approach. In the first stage of division, we consider both sequential and parallel approaches for simplifying the task into a set of co-sparse unit-rank estimation (CURE) problems, and establish the statistical underpinnings of these commonly-adopted and yet poorly understood deflation methods. In the second stage of division, we innovate a contended stagewise learning technique, consisting of a sequence of simple incremental updates, to efficiently trace out the whole solution paths of CURE. Our algorithm has a much lower computational complexity than alternating convex search, and the choice of the step size enables a flexible and principled tradeoff between statistical accuracy and computational efficiency. Our work is among the first to enable stagewise learning for non-convex problems, and the idea can be applicable in many multi-convex problems. Extensive simulation studies and an application in genetics demonstrate the effectiveness and scalability of our approach.

報告人簡介：鄭澤敏，現(xiàn)為中國科學技術大學管理學院教授、統(tǒng)計與金融系主任、博士生導師，其研究方向是高維統(tǒng)計推斷和大數(shù)據(jù)問題。鄭澤敏博士在橫跨這一領域的若干關鍵研究課題上取得了富有創(chuàng)造性的研究成果，研究成果發(fā)表在Journal of the Royal Statistical Society: Series B (JRSSB )、Annals of Statistics (AOS)、Operations Research(OR)、Journal of Machine LearningResearch(JMLR)、Journal of Business & Economic Statistics (JBES)等國際統(tǒng)計學、機器學習、計量經(jīng)濟學及管理優(yōu)化領域的頂級期刊上，曾獲南加州大學授予的優(yōu)秀科研獎和美國數(shù)理統(tǒng)計協(xié)會頒發(fā)的科研新人獎,并于2017年入選中組部青年創(chuàng)新人才計劃。個人主頁：http://bs.ustc.edu.cn/chinese/profile-302.html

會議信息：

黃磊 邀請您參加騰訊會議

會議主題：統(tǒng)計學系列學術報告

會議時間：2022/05/13 08:30-11:30 (GMT+08:00) 中國標準時間 - 北京

點擊鏈接入會，或添加至會議列表：

https://meeting.tencent.com/dm/72GXrs9KE0W9

騰訊會議：187-336-030

會議密碼：220513