Seminar| Institute of Mathematical Sciences
Time: Thursday, January 4th , 2024 , 16:00-17:00
Speaker: Menggang Yu，Department of Biostatistics University of Michigan School of Public Health
Abstract: The multiple instance learning (MIL) literature, where instances of data are naturally grouped into collections (“bags”) with a label, but where the labels on the instances themselves are unobserved, has primarily focused on binary classification. We consider modeling ordinal outcomes in the MIL framework. We explore many combinations of popular and applicable approaches through a large statistical experiment designed to detect their performance across data sets from several problem applications. From leveraging the findings of this experiment, we gain additional insight into a motivating breast cancer biomarker application.