Models and Methods for Random Networks

Evolution and Dynamics 1: Decentralized Network Navigation

Course Materials

Project

Deadline

  • Homework 3 due today.

Illustrations

These animations shows the evolution of the decentralized navigation algorithm (red) on a square 2-dimensional grid with shortcuts, as described in Jon Kleinberg's paper "The Small-World Phenomenon: An Algorithmic Perspective". In parallel, the green path shows the shortest path that a centralized algorithm would choose.

  • Scale-free shortcuts. The shortcut exponent is set to 2 in this example, the unique value for which efficient decentralized navigation is possible in this model. This holds essentially because the navigation algorithm can find useful shortcuts at every scale quite efficiently. The centralized algorithm is performing slightly better because it "sees" all the shortcuts in the network, not just those of visited nodes.

  • Shortcuts too local. The shortcut exponent is set to 4 in this example, which means that shortcuts tend to be very local w.r.t. the grid. The decentralized routing algorithm can only make slow progress.

  • Shortcuts too global. The shortcut exponent is set to 0 in this example, which means that each shortcut goes to a node sampled uniformly at random from the set of all nodes. The decentralized routing algorithm makes fast progress first, but slows down when it approaches the destination, because no "local shortcuts" can be found. Even though a short path exists, it cannot be discovered by the local navigation algorithm.

[prev] [back] [next]