講座：Stochastic Optimization Forests
題 目：Stochastic Optimization Forests
演講人：毛小介 博士生 美國康奈爾大學
主持人：李成璋 助理教授 上海交通大學安泰經濟與管理學院
We study contextual stochastic optimization problems, where we leverage rich auxiliary observations (e.g., product characteristics) to improve decision making with uncertain variables (e.g., demand). We show how to train forest decision policies for this problem by growing trees that choose splits to directly optimize the downstream decision quality, rather than splitting to improve prediction accuracy as in the standard random forest algorithm. We realize this seemingly computationally intractable problem by developing approximate splitting criteria that utilize optimization perturbation analysis to eschew burdensome re-optimization for every candidate split, so that our method scales to large-scale problems. We prove that our method achieves asymptotic optimality in decision making. We validate our method empirically in portfolio optimization and inventory management problems, demonstrating the value of optimization-aware construction of forests and the success of our efficient approximations. We show that our approximate splitting criteria can reduce running time hundredfold, while preserving superb decision-making performance.
Xiaojie Mao is a fifth-year PhD student in the Department of Statistics and Data Science at Cornell University. He works on data-driven decision making and causal inference, with a particular focus on developing flexible machine learning methods for problems arising in these areas. He has published in Management Science and top machine learning conferences, and his research work is in the finalists for the Applied Probability Society Best Student Paper Competition 2020. Prior to his graduate study, he was an undergraduate student majoring in Mathematical Economics at Wuhan University. See more details at https://xiaojiemao.github.io/.