Swarm Intelligence

simultaneously indirectly evolving network architecture was presented.

The methodology seems to overcome the first four disadvantages to pre-

vious approaches listed earlier. With respect to the fifth “disadvantage,”

our work on this and other projects leads us to agree with Hancock’s con-

clusion that permutations are more a help than a hindrance and, further,

that multimodality is one reason why particle swarm optimization (and

other evolutionary algorithms) works so well. Classification of the Iris

Data Set (both normalized and unnormalized) was used to show that

good results can be obtained by evolving weights and architecture of

neural networks using ...