|
Another way of analyzing the genotypes that have given good performance is to look at the commonalities of high-fitness networks. By repeating the experiment, this time penalizing genotypes that use what was common for good networks, I will be able to measure how important these characteristics were; if the characteristics are important enough, the system should use them again, even at the expense of some penalty to the genotypes that use them. For example, if a high number of hidden nodes end up using a product transfer function in the original experiment, the experiment will be repeated, but this time a penalty will be assigned to networks that choose to use this type of function.
|
Back to the previous subject
Back to the Table of Content
To the next subject