Outperformance Inverse Optimization: Learning Objective Functions that Outperform Agent Decisions
Outperformance inverse optimization learns MILP objective weights that beat observed decisions on every component.
Standard inverse optimization for mixed-integer linear programs seeks objective weights that reproduce observed decisions as optimal, which produces compromise weights when those decisions are suboptimal. Outperformance inverse optimization instead seeks weights whose optimal solution beats the observed action in every component at each state. The proposed loss can be evaluated with forward-problem oracles alone and is minimized with gradient-based and DC algorithms. When the outperforming solution is unique, generalization error is bounded inversely with the number of observations, and experiments on synthetic and real data improve prediction of outperforming solutions.
- Learns weights whose optima beat observed actions on every component
- Loss uses only forward MILP oracles, with gradient and DC methods
- Generalization error falls inversely with the number of observations
- Synthetic and real experiments beat prior inverse-optimization methods
From the original article
1 min read at arxiv.orgInverse optimization estimates the weights of an objective function that explain observed decisions as optimal solutions, and is used in a variety of fields. For mixed-integer linear programs (MILPs), existing methods aim to reproduce the observations as optimal solutions, and thus learn compromise weights when the observations are suboptimal.
By Akira Kitaoka · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.