Information Acquisition and Representation Methods for Real-Time Asset Management

摘要

The last three years have been exceptionally productive. Our research focused on two complementary themes: optimal learning, which addresses the efficient collection of information, and approximate dynamic programming, which is a modeling and algorithmic strategy for solving complex, sequential decision problems. These problems arise in the control of complex machinery, R&D portfolio optimization, materials science (sequential design of experiments), communications, and a wide range of resource allocation problems that arise in operations and logistics including mid-air refueling, spare parts management, emergency response, and robust allocation of fuel, medical supplies and food. In the process of making advances in approximate dynamic programming, we found ourselves making contributions to an area that is proving to be critical to both lines of investigation: machine learning. In fact, we have come to realize that machine learning is starting to play a critical role in the advancement of our ability to solve complex stochastic programming problems, and it began to play an important role both in optimal learning and approximate dynamic programming.