Synaptic weight evolution.

Simulations use K = 4 and have a switch-point after 500 stimuli. The weights are averages over 100 simulations using the same rules and over the intrinsic redundancy of k-hot representations. A Evolution of all the weights from stimulus 1 to the prediction error population P1. The weights of the four possible transitions increase their magnitude while the others decay, until the change point. From the evolution of the weights we can observe that rule 1 has 2 transitions in common with rule 2. B Same for P2. C Same but averaged over the two populations. D Same as before, but all weights for all the stimuli.

(EPS)