报告题目：Two Head Neural Network based on Maximum Rank Correlation Loss 

报告人：张三国 教授（中国科学院大学） 

报告时间：2024年6月25日（周二）下午16：00-17：00 

报告地点：葡京新集团350入口235报告厅 

报告摘要: Recent years, deep learning has been received much attention due to powerful performance in many areas. Exploiting a large amount of labeled data to train is one typical reason of its success. In data sparsity problem, multi-task learning(MTL) is a good recipe by exploring useful information from other related learning tasks. We study two-head neural network including typical classification an regression tasks with high dimension input. Maximum Rank Correlation(MRC) is a measure of correlation between the ranking of true labels and fitted labels. It is invariant with respect to the functional forms of prediction and error distribution, which can deal with different scale of classification and regression tasks. We apply MRC loss, which is robust to error distribution, to two-head neural network. High computational cost and difficulty to adapt to regression task is discussed in this paper. Variable selection effect with high dimension input and model performance will be developed in experiments. 

报告人简介：张三国，中国科学院大学数学科学学院教授、博士生导师，2002年毕业于中国科学技术大学，获博士学位。先后与03年8月-04年7月在香港中文大学统计学系，07年2月-08年8月在美国范德堡大学（Vanderbilt University）的医学中心公众健康研究所与生物统计系从事博士后研究工作。多年来一直从事高维数据分析，生物与医学统计、统计机器学习教学科研工作，曾获得2017年中国科学院优秀导师奖。近五年来发表论文三十余篇，相关研究成果发表在Sciences in China-Mathematics, JASA，Bioinformatics, Biometrics等数理统计、生物统计和生物信息学领域的权威期刊。主持多项纵向和横向课题，包括国家自然科学天元基金重点、面上、青年项目，企业和军工科研项目等。