I ran a test training the MLP with data filtered such that at least one cluster must change state between subsequent frames. This time, the result is a periodic dream that does not stabilize over 10,000 iterations: On a hunch I figured the network state would be highly dependent on initial conditions, which is true. […]
↧