Robust Stochastic Optimization Made Easy with RSOME
Operations Research and Operations Management
Zhi Chen, Melvyn Sim, Peng Xiong
Published in Management Science, May 2020
Uncertainty, due to inevitable data contamination, estimation errors, as well as model misspecification, is ubiquitous in real-world problems. As a consequence, decision makers, both individuals and organizations, frequently need to make judgements based on incomplete information. This may concern crucial areas such as future customer demand, raw material prices, or exchange rates. The absence of definitive data sets has driven a rapid growth in mathematical paradigms for generic modelling and optimization under uncertainty over the past few decades.
Of these paradigms, robust optimization is arguably one of the most popular and powerful and its successful applications have appeared in a wide spectrum of decision-making problems under uncertainty, ranging from supply chain and logistics, revenue management, and operations management, to machine learning and engineering. Parallel to theoretical advances in robust optimization, algebraic modelling packages, as accompanying technology to facilitate modelling robust optimization problems in practical use, have been developed on various platforms for scientific computing, including C, C++, and MATLAB.
In a recent fast-track paper “Robust Stochastic Optimization Made Easy with RSOME” published in Management Science in May 2020, Zhi Chen, Assistant Professor in the Department of Management Sciences, and his collaborators from the Department of Analytics & Operations, National University of Singapore, introduce a new robust optimization model called robust stochastic optimization (RSO), which unifies many existing robust models for prescribing decisions when facing uncertainty.
The RSO model also opens up new approaches, including those inspired by machine learning techniques. To facilitate the prototype, testing, and implementation of RSO models for practical applications in an intuitive and efficient manner, the authors develop a new algebraic modelling package RSOME. The package has the potential to help practitioners in navigating and evaluating the plethora of approaches, and to address a wide variety of uncertainty-affected optimization problems in practice. For example, a team of researchers from the National University of Singapore has used the RSO model to decide vehicle pre-allocation under uncertain weather conditions, for which the optimal decisions are obtained with the aim of the RSOME package.
