Design of fault tolerant networks with agent-based simulation of Physarum polycephalum

Abstract

In this work we evaluate slime mold inspired algorithms, that gained a lot of attention in renowned journals recently, for their ability to construct fault tolerant connection networks. In previous work, experiments with a real slime mold Physarum polycephalum as well as computer simulations based on a tube model of the slime mold showed, that the slime mold (and simulations thereof) is able to construct fault tolerant and efficient transport networks similar to the actual Tokyo rail system [1]. However the quality of the solutions of the real slime mold show big variations, and the tubular computer simulation does not seem to reproduce the natural slime mold very well, since the constructed networks do not show the variety of the naturally build ones, instead they show a heavy dependence of one simulation parameter. Thus in our work we present a different approach for construction of fault tolerant connection networks for the Tokyo rail system using an agent based simulation of Physarum polycephalum. Analysis of the results show that the agent based simulation reproduces the variance in the behavior of the natural slime mold much better. Analyzing the cost benefit ratio of bio-inspired network construction we however conclude that it might be worth to consider classical efficient computational algorithms for the problem of constructing minimal fault tolerant networks.