Experiment 4 is from 2001 Tech Report pg. 5. This is a 20x20 nets, and is presented two types of patterns, neurons from the top half or neurons from the bottom half, but the middle two rows appear in both. Uses ParametersExp4.dat
When this is started up, a new random net is generated. To look at the synaptic weights, go to the measure menu and select Print Synaptic Variance. This prints the average weight from one group of neurons to another. In this case, group 0 is the top 9 rows, group 1 is the bottom 9, and group 2 is the middle 2 rows.
Initially, this shows that all synaptic weights are almost the same, as is expected in a randomly generated net. As the system runs, the weights rapidly approach those predicted by the correlatory learning rule.
If the learning rate is changed to .05, the weights approach these correlatory values more slowly. Similarly, if a more reasonable presentation mechanism is used (by turning Change Pattern Each Time off), these weights will take longer to converge, and there will be more variance.