学术报告:逻辑回归模型中因素间交互关系的探测

发布者:潘春德发布时间:2018-01-23浏览次数:185

报告题目:逻辑回归模型中因素间交互关系的探测

间:2018125日(周四)上午10:00

告人:徐力,密歇根大学 (University of Michigan)

点:中国矿业大学文昌校区信电楼400

摘要: In the fields of machine learning and big data analytics, multiple factors and their interactions have been considered to play critical roles in determining the outcome. Identifying the interactive effects among features can help to make outcome prediction and understand the mechanization of system behavior. In this talk, we first show how to detect the pairwise interactions among a set of binary covariates in logistic regression from a limited number of samples. We propose an algorithm that is based on a maximum-weight spanning tree with respect to the plug-in estimates of certain quantities not only has strong theoretical performance guarantees, but can also outperform generic feature selection algorithms for detecting the pairwise interactions. Next, we study how the multivariate synergy, an information theoretical measure, helps to make feature selection with high-order interactions. It is shown that the multivariate synergy roughly relies on a small subset of the interaction parameters, sometimes on only one interaction parameter. We further establish the rigorous theoretical analysis on the error estimation.

报告人简介:

徐力博士,作为黑龙江省牡丹江市高考理科状元进入北京大学, 获得计算机科学与技术学士学位、数学与应用数学双学士学位。 2010年至2014年间在香港大学(University of Hong Kong)获得数学博士学位。之后于美国德州农工大学(Texas A&M University)从事博士后研究工作。曾作为访问学者访问美国密歇根大学、密歇根数据科学中心、香港中文大学(深圳)、深圳市大数据研究院等学术机构。现在是美国密歇根大学(University of Michigan)博士后。徐力博士的主要研究领域是机器学习、信息论及大数据分析。