Biological Computation

Artificial Neural Networks    ◾    193

we identied in the set of strings that was memorized. In other words, as

the weights reect the types of relations between the bits, the network has

succeeded in learning the rule governing the samples. Obviously, regular-

ity can also be learned by the feedforward networks described earlier.

Hopeld networks can also be used for optimizations. We saw in the

proof of the convergence of the retrieval process that a set of weights corre-

sponds to an energy function. is function is minimized by the process of

updating the values of the neurons such that at the end of the process neu-

rons connected by an edg ...