Distributionally Strong Optimization with Determination-Dependent Data Discovery
Authors: Qing Jin, Angelos Georghiou, Phebe Vayanos, Grani A. Hanasusanto
Summary: We research two-stage distributionally strong optimization (DRO) issues with decision-dependent data discovery (DDID) whereby (a portion of) the unsure parameters are revealed provided that an (usually expensive) funding is made within the first stage. This class of issues finds many necessary purposes in choice issues (e.g., in hiring, challenge portfolio optimization, or optimum sensor location). Regardless of the issue’s huge applicability, it has not been beforehand studied. We suggest a framework for modeling and roughly fixing DRO issues with DDID. We formulate the issue as a min-max-min-max downside and undertake the favored Ok-adaptability approximation scheme, which chooses Ok candidate recourse actions here-and-now and implements the most effective of these actions after the unsure parameters that had been chosen to be noticed are revealed. We then current a decomposition algorithm that solves the Ok-adaptable formulation precisely. Specifically, we devise a reducing airplane algorithm which iteratively solves a relaxed model of the issue, evaluates the true goal worth of the corresponding answer, generates legitimate cuts, and imposes them within the relaxed downside. For the analysis downside, we develop a branch-and-cut algorithm that provably converges to an optimum answer. We showcase the effectiveness of our framework on the R&D challenge portfolio optimization downside and the most effective field downside.
