报告题目：Large Class Classification with Application in Digital Economy
报 告 人：王思鉴 罗格斯大学
报告时间：2023年6月13日(周三) 14:00-15:30
报告地点：中国科学院大学中关村校区教学楼S406
腾讯会议：247-366-522
内容摘要
In the realm of digital economy, challenges often arise in tackling multi-class classification problems with a vast number of classes, commonly found in tasks like product categorization and user behavior analysis. Traditional classification methods struggle to deliver satisfactory results on these large-scale problems due to their computational demands and subpar performance. This talk focuses on viewing classification from a class-embedding perspective, where classification is seen as balancing the forces acting upon the class-embedding vector from both correctly and incorrectly classified data points. From this viewpoint, we propose a framework that introduces an adaptively weighted loss function designed to handle such complex classification tasks more efficiently. We also utilize sampling techniques to expedite computation. The utility and effectiveness of these methods are showcased through simulation studies and real-world data analysis, underscoring their potential for improvements in digital economy applications.
主讲人简介
Dr. Sijian Wang obtained his BS in Mathematics from Tsinghua University and has a Ph.D. in Biostatistics from The University of Michigan. He was assistant professor (tenure track) and associate professor (with tenture) in the Department of Statistics and Department of Biostatistics & Medical Informatics at University of Wisconsin, Madison. He is currently an associate professor (with tenure) in the Department of Statistics, a residence member of Institute for Quantitative Biomedicine and Co-Director of Financial Statistics & Risk Management and Co-Director of Data Science at Rutgers University. His research interests include statistical/machine learning and statistical modeling.
