Conic Reformulations for Kullback-Leibler Divergence Constrained Distributionally Robust Optimization and Applications

Burak Kocuk

DOI: https://doi.org/10.11121/ijocta.01.2021.001001

2024-11-11

Abstract:In this paper, we consider a distributionally robust optimization (DRO) model in which the ambiguity set is defined as the set of distributions whose Kullback-Leibler (KL) divergence to an empirical distribution is bounded. Utilizing the fact that KL divergence is an exponential cone representable function, we obtain the robust counterpart of the KL divergence constrained DRO problem as a dual exponential cone constrained program under mild assumptions on the underlying optimization problem. The resulting conic reformulation of the original optimization problem can be directly solved by a commercial conic programming solver. We specialize our generic formulation to two classical optimization problems, namely, the Newsvendor Problem and the Uncapacitated Facility Location Problem. Our computational study in an out-of-sample analysis shows that the solutions obtained via the DRO approach yield significantly better performance in terms of the dispersion of the cost realizations while the central tendency deteriorates only slightly compared to the solutions obtained by stochastic programming.

Optimization and Control

What problem does this paper attempt to address?

This paper attempts to address the problem of how to handle optimization problems with Kullback-Leibler (KL) divergence constraints in Distributionally Robust Optimization (DRO). Specifically, the paper considers a DRO model where the uncertainty set is defined as the set of distributions whose KL divergence from the empirical distribution is bounded. By leveraging the fact that KL divergence is an exponential cone representable function, the authors transform the robust dual form of the DRO problem with KL divergence constraints into a dual exponential cone constrained programming problem. This transformation allows the original optimization problem to be directly solved using commercial cone programming solvers.

The main contributions of the paper include:
1. **Problem Modeling**: Proposing the DRO problem with KL divergence constraints and providing its robust dual form.
2. **Cone Programming Transformation**: Transforming the problem into a dual exponential cone constrained programming problem through cone duality.
3. **Application Examples**: Applying the general model to two classic optimization problems—the Newsvendor Problem and the Uncapacitated Facility Location Problem.
4. **Numerical Experiments**: Validating the performance of the proposed method in out-of-sample analysis through numerical experiments, particularly in terms of cost dispersion and risk metrics improvement.

The structure of the paper is as follows:
- **Introduction**: Introduces the two main paradigms under decision uncertainty—stochastic programming and robust optimization, and introduces the concept of Distributionally Robust Optimization (DRO).
- **Preliminaries**: Reviews the basic concepts of convex analysis and probability theory, particularly the relevant theories of cone programming and KL divergence.
- **Main Results**: Details the robust dual form of the DRO problem with KL divergence constraints and its transformation into cone programming.
- **Application Examples**: Discusses the DRO versions of the Newsvendor Problem and the Uncapacitated Facility Location Problem, providing specific models and numerical experiment results.
- **Conclusion**: Summarizes the main findings of the research and future research directions.

Through this work, the paper demonstrates that the DRO problem with KL divergence constraints can be effectively solved using cone programming techniques and shows good performance in practical applications.