Generative Robust Optimisation

The article introduces Generative Robust Optimisation (GRO), a novel framework that utilizes deep generative models to define uncertainty sets, allowing for the representation of complex dependencies in real-world data. It employs a Wasserstein Adversarial Autoencoder with Gaussian mixture model-guided training and constraint-consistency regularization, and evaluates its performance using a five-point framework. This approach enhances the robustness of optimization under uncertainty by producing expressive, well-calibrated, and computationally tractable uncertainty sets, which is critical for practitioners dealing with complex optimization problems in AI applications.