Network Pruning and Growth - Probabilistic Optimization

Abstract

Being the backbone of many human-made systems, networks require both pruning and growth to adapt to changing demand. We develop a message passing-based framework for analyzing and addressing the two-level optimization problem of edge removal/addition for indirectly-dependent objectives. As exemplar problem we use routing in optical communication networks to minimize capability loss (removal) or maximize capacity (addition). The methods developed result in lower path-lengths and higher capacity topologies with respect to existing ones and are suitable for a broad range of network design tasks.