Computational Methods for Stochastic Optimization and Machine Learning

3 credits · College of Engineering catalogue

Advisory Prerequisite: Optimization at the level of IOE 410 (or its equivalent) as well as undergraduate probability. Enforced Prerequisite: Graduate Standing. (3 credits) Offers a comprehensive understanding of computational methods for stochastic optimization and machine learning (ML). Subject matter is divided into three parts: (I) Continuous optimization; (II) dynamic programming; and (III) reinforcement learning. Topics include optimization algorithms (descent methods, linesearch methods, momentum methods, stochastic gradient methods); dynamic programming (infinite-horizon problems, value iteration, and policy-iteration; and reinforcement learning (RL) (multi-armed bandits, TD learning, and Q-learning). Python-based computational implementations and subsequent applications to applications in ML and RL will be emphasized.