- This event has passed.
Quantitative Issues in Cancer Research Working Seminar
September 28 @ 1:00 pm - 2:00 pm
Cathy WangDoctoral Student, Department of Biostatistics, Harvard University”Multi-Study Boosting” ABSTRACT: Training prediction models on multiple studies is more likely to produce cross-study replicable results. Recent work in multi-study learning leverages ensembles of prediction models, each trained on one, or a subset, of the studies, to design cross-study replicable prediction algorithms. Boosting is an ensemble learning technique that sequentially combines weak learners to produce a strong learner. We extend boosting to the multi-study setting by training a weighted sum of trees as the base learner and rewarding cross-study replicability with stacked regression weights during the training phase. We assess the empirical performance of multi-study boosting with simulations.