Visualising Evolutionary Computation

In terms of visualisation three main approaches can be identified: Firstly, tracing the ancestry of the evolving individuals or representing the genetic composition of individuals (Collins, 2002; Hart and Ross, 2001; Smith et al., 2002). The information presented in this approach is of interest to researches in Evolutionary Computation, rather than designers or users of such systems.

Secondly, the simplest attempt to introduce visualisation to help designers was that of a simple fitness plot and later on interaction with the fitness plot to backtrack the evolution process (Mathew,

2000) , and in the multi-objective case, plotting the Pareto front or Pareto surface (Grierson and Khajehpour, 2002), puts this technique to good effect.

Thirdly, in high dimensional spaces it is difficult for the designer to see relationship between the design variables or the interaction with the objectives. The work of Abraham and Parmee (2004) uses a box-plot technique to display design variable – objective interaction, whilst Hayashida and Takagi (2000) used techniques to map high dimensional spaces to a 2-D mapped space.

DISCUS: Distributed Innovation and Scalable Collaborations in Uncertain Settings.

In all cases discussed so far, visualisation is performed during or after the GA run has terminated and the user is not able to interact with the system either to refine the search or to move the search to a different area of the search space.