时间： 2020年1月6日 16：00-17：00
题目：Modeling and Active Learning for Experiments with Quantitative-Sequence Factors
A new type of experiments which target on finding optimal quantities of a sequence of factors is drawing much attention in medical science, and are also encountered in engineering, chemistry, physics, managements, food science and other disciplines. Traditionally, researchers only focus on either optimizing the values for factors or identifying the optimal sequences to arrange them. In this paper, we consider the simultaneous optimization for both the quantity and sequence, which is defined as a new type of factors: Quantitative-Sequence (QS) factors. Due to the extremely large and non-continuous solution spaces for such experiments, it is of great challenge to identify optimal (or near optimal) settings via only a few runs. To tackle this problem, we propose a novel active learning approach (QS-learning) which consists of three components: a novel Mapping-based Additive Gaussian Process (MaGP) model, an intelligent sequential scheme (QS_EGO) and a new class of optimal experimental designs (QS-design). We illustrate the superiority of the proposed method via a real drug experiment on lymphoma and several simulation studies with interpretable backgrounds.
Qian Xiao is Assistant Professor at Department of Statistics, University of Georgia. He received his PhD degree in Statistics at University of California, Los Angeles in 2017. His research interests include computer experiments, experimental designs, drug combination experiments and modeling (Kriging model, sparse model).