Next: 2 Even Parity
Up: 1 Artificial Ant
Previous: 2 Ant Fitness Assessment
The Artificial Ant problem was investigated in numerous studies.
and Poli (2002) examined several aspects of the
search space for the Ant problem. The authors measure the search effort
required for genetic programming, several variations of genetic programming, several random search
techniques and several stochastic search methods. While genetic programming performed
at least as good as the random search methods, some versions of
stochastic hill-climbing, including population based hill-climbing, performed
equally well or better than genetic programming.
Langdon and Poli noted that the Ant problem is highly deceptive. That is,
there are many possible solutions that are symmetrical, and because
there is no requirement that a particular path is followed, solutions
with equal fitness can take very different approaches to solving the
Langdon and Poli later showed that by encouraging
the population to follow similar paths, the problem became considerably
easier for genetic programming.