Figure 1: Snapshots from the evolution of an addition-network, including the final result at the very bottom. Node X1 is not needed for the computation and could be regarded as “junk“.
Figure 2: Development of the best, average and worst fitness during a typical run.
Figure 3: An example for file
buildingblocks.txt
Figure 4: An example file
objectivefunction.txt
Figure 5: An example file
options.txt
Figure 6: A network resulting from the evolution of an addition. Although it looks more complicated than a hand-designed network, it perfectly performs its task.
Figure 7: Outline of the two-level evolutionary algorithm
Figure 8: Comparison of the average fitness development for the six strategies depicted in the text. Left column: two-level EA, middle column: one-level for many generations, right column: one-level with a large population. Upper row is without AIC, lower row uses AIC.
