The different rewards adminstered to connections by our learning rules, were set to the following values:
These values were chosen so as to reflect the perceived relative importance of a certain learning rule to the network development. Clearly Frequency, as the sole selection principle for existing connections, was considered most important and was therefore equipped with a relatively high reward. Since Transitivity and Symmetry both guarantee the generation of new and original links whose usefulness has not yet been proven, they were thus equipped with relatively smaller rewards.
- Frequency: Fb=1
- Transitivity: Tb=0.5
- Symmetry: Sb=0.3
We presumed the succes of our experiment not to be critically dependent upon the precise reward values involved.