The paper discusses extending the Predict-Then-Optimize (PtO) methodology to optimization problems with nondifferentiable Ordered Weighted Averaging (OWA) objectives. OWA objectives ensure fairness and robustness in decision models. The authors propose new training techniques and application settings to integrate optimization of OWA functions with parametric prediction effectively. This extension allows for fair and robust optimization under uncertainty, a significant advancement in artificial intelligence and machine learning.
Publication date: 13 Feb 2024
Project Page: https://arxiv.org/abs/2402.07772v1
Paper: https://arxiv.org/pdf/2402.07772
