Slide 16 of 45
Notes:
Results consistent with understanding of algorithm:
a too low => too little importance associated with the trail. High a means that that trail is *very* important and therefore ants tend to choose edges chosen by other ants in the past (too little exploration of the search space).
Optimal values determined experimentally a =1, b=5, r=0.5, Q=100.
Within range of parameter optimality, ant-cycle always finds good solutions for all tested problems.
The algorithm quickly finds good solutions (when compared to say, GAs) and does not exhibit stagnation behavior -- the ants continue to look for new and better solutions.
AS system sensitivity investigated w.r.t. problem dimensionality. Found little sensitivity with increasing problem size.
No theory currently to explain parameter settings. Parameters need hand crafting.
