题目：Multi-label learning using l1-norm penalty

报告人：刘华文（浙江师范大学副教授，NLPR开放课题客座）

主持人：刘成林

时间：7月31日上午10:30

地点：智能化大厦三层第二会议室

摘要：
  Multi-label data is prevalent in practice and multi-label learning has now received more and more attentions due to its great potential applications. However, there are two major challenges the multi-label learning algorithms will encounter. One is the correlation of labels and the other is the high dimensionality of data. In this talk, we will introduce two effective learning algorithms for high-dimensional multi-label data, which can not only capture the label correlations, but also perform dimensionality reduction simultaneously. The first one exploits logistic regression to train models for multilabel data. Moreover, an elastic net penalty is involved to regularize the logistic regression, aiming at dealing with the problems raised from high dimensionality. The second one adopts partial least squares regression regularized with $l_1$-norm penalty, so as to discover the relevance between the input space and the output space. The experimental results conducted on public data sets show that our method achieved better performance than popular multi-label classifiers in most cases.

个人简历
  刘华文，博士，副教授，2010年获吉林大学计算机软件与理论博士学位（博士学位论文获吉林省优秀博士论文）。2010年加入浙江师范大学计算机科学系。2011年9月，赴澳大利亚南澳大学计算机与信息系统学院访学。主要研究方向为