Opening Bottlenecks on Weighted Networks by Local Adaptation to Cascade Failures
Jeff Alstott, University of Cambridge and US National Institutes of Health
Complex systems are often structured as weighted networks, with system components connected to each other via links that have some weight. The ability for activity or information to cascade or flow through the network will be determined by the positioning and weights of these links. Network elements with a low ability to propagate activity are bottlenecks, terminating activity. A network with no bottlenecks can transfer activity more effectively along arbitrary sets of network elements. Here we model activity cascades on complex networks to identify the typical temporal profile of activity cascades, in which activity moves from high out strength nodes to low strength bottlenecks. We take advantage of that temporal profile with a simple adaptation process, which alters link weights in response to the timing of intrinsic cascading activity. This adaptation process removes bottlenecks and increases the accessibility of all paths on the network. This new state can be a useful starting point for a system to adapt to new inputs, such as a young brain learning new associations.