Look-Ahead Robust Network Optimization with Generative State Predictions

Fei Xu Yu, Zuyuan Zhang, Emily Grob, Gina Adam, Sean Coffey, Nathaniel D. Bastian, Tian Lan

Keywords: Network Optimization, Robust optimization, Distributed robust optimization, Diffusion models

Abstract: In network optimization, especially in environments with significant uncertainty, traditional approaches are often overly conservative which leads to inefficiencies. Robust network planning is essential when network conditions can vary unpredictably. Traditional robust approaches tend to protect against all types of uncertainty, including those with minimal impact, resulting in suboptimal performance and resource allocation. To address this, we propose leveraging diffusion models to generate future network states based on historical data, capturing realistic variations without overgeneralizing uncertainty. Our method defines an uncertainty set based on these generated states, focusing on probable scenarios rather than extreme outliers. Using this uncertainty set, we use robust optimization to allocate resources and ensure network reliability under dynamic conditions. Preliminary experiments demonstrate that our approach achieves a balance between robustness and efficiency, significantly outperforming rational methods in realistic network scenarios.

Submission Number: 5