I'm relatively new to neural networks, and I've been trying to program my own Hopfield network. I got it to the point where it can reliably reproduce a single pattern from a completely scrambled starting state, for up to 400 neurons (potentially more, my computer takes a while for anything bigger than that). When I try it with more than 1 stored pattern however, it appears to settle on a spurious state that looks like some combination (sum?) of them rather than any of them in particular.
My question is whether the method I'm using to generate my patterns, the size of the network, or any conditions in general can cause this to happen. At the moment, I'm using randomly-generated patterns. To test the network, I set it to be identical to one of the patterns, introduce some random noise by flipping a certain number of nodes, and then seeing if it can recreate the pattern. Could anything about that process be causing the network to settle on that combination of states?
Thanks in advance.