Abstract:We study two-stage distributionally robust optimization (DRO) problems with decision-dependent information discovery (DDID) wherein (a portion of) the uncertain parameters are revealed only if an (often costly) investment is made in the first stage. This class of problems finds many important applications in selection problems (e.g., in hiring, project portfolio optimization, or optimal sensor location). Despite the problem's wide applicability, it has not been previously studied. We propose a framework for modeling and approximately solving DRO problems with DDID. We formulate the problem as a min-max-min-max problem and adopt the popular K-adaptability approximation scheme, which chooses K candidate recourse actions here-and-now and implements the best of those actions after the uncertain parameters that were chosen to be observed are revealed. We then present a decomposition algorithm that solves the K-adaptable formulation exactly. In particular, we devise a cutting plane algorithm which iteratively solves a relaxed version of the problem, evaluates the true objective value of the corresponding solution, generates valid cuts, and imposes them in the relaxed problem. For the evaluation problem, we develop a branch-and-cut algorithm that provably converges to an optimal solution. We showcase the effectiveness of our framework on the R&D project portfolio optimization problem and the best box problem.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to conduct two - stage distributionally robust optimization (DRO) under decision - dependent information discovery (DDID). Specifically, the paper focuses on how to effectively model and solve such problems when the investment decisions made in the first stage can reveal some uncertain parameters. Such problems are very common in practical applications, such as R & D project portfolio optimization, optimal sensor location selection, and selection problems. In these scenarios, uncertain parameters can only be observed after corresponding investments or measurements have been made, so the actions of decision - makers directly affect the time of information discovery. The main contributions of the paper include: 1. **Problem Modeling**: Propose an adaptive distributionally robust optimization model that takes into account decision - dependent non - anticipativity constraints and prove that this model can be transformed into an equivalent min - max - min - max robust optimization problem. 2. **Approximation Method**: Adopt the K - adaptability approximation scheme to deal with this complex optimization problem. By pre - selecting K candidate strategies and implementing the optimal one after some of the uncertain parameters are revealed, a trade - off between computational complexity and solution quality is achieved. 3. **Solution Algorithm**: Develop a decomposition algorithm to accurately solve the K - adaptability problem. This algorithm iteratively solves a relaxed version of the problem, evaluates the true objective value of the current solution, generates valid cutting planes, and adds them to the relaxed problem until the optimal solution is found. For the evaluation problem, a branch - and - cut algorithm is also developed to ensure convergence to the optimal solution. 4. **Numerical Experiments**: Demonstrate the effectiveness of the proposed method by studying instances of R & D project portfolio optimization and the optimal box problem. The experimental results show that the decomposition algorithm outperforms existing solvers in solving large - scale problems, and the distributionally robust optimization solution improves the performance by an average of 70% compared to the pure robust optimization solution. In conclusion, this paper provides a framework for modeling and solving distributionally robust optimization problems in the context of decision - dependent information discovery, which has important theoretical and practical significance.

Distributionally Robust Optimization with Decision-Dependent Information Discovery

Multistage Distributionally Robust Optimization for Integrated Production and Maintenance Scheduling

Distributionally Robust Decision Making Leveraging Conditional Distributions

Distributed Distributionally Robust Optimization with Non-Convex Objectives

Wasserstein Distributionally Robust Optimization with Heterogeneous Data Sources

Efficient Algorithms for Distributionally Robust Stochastic Optimization with Discrete Scenario Support

Revealing Decision Conservativeness Through Inverse Distributionally Robust Optimization

Distributionally robust optimization for planning and scheduling under uncertainty

Distributionally Favorable Optimization: A Framework for Data-Driven Decision-Making with Endogenous Outliers

Distributed Robust Optimization in Networked System.

Bounds for Multistage Mixed-Integer Distributionally Robust Optimization

Distributionally Robust Optimization with Multimodal Decision-Dependent Ambiguity Sets

Stochastic First-Order Algorithms for Constrained Distributionally Robust Optimization

Distributionally Robust Multi-Stage Decision Problem and Satisficing

Distributionally Robust Optimization under Distorted Expectations

Decision bounding problems for two-stage distributionally robust stochastic bilevel optimization

Federated Distributionally Robust Optimization with Non-Convex Objectives: Algorithm and Analysis

Numerical method for approximately optimal solutions of two-stage distributionally robust optimization with marginal constraints

Optimized Dimensionality Reduction for Moment-based Distributionally Robust Optimization

Distributionally Risk-Receptive and Robust Multistage Stochastic Integer Programs and Interdiction Models

Differentiable Distributionally Robust Optimization Layers