§11 Associative Memory — Hopfield Network
wij = (1/N) · Σμ ξiμ · ξjμ
E = -(1/2) · Σi,j wij · si · sj
E = -(1/2) · Σi,j wij · si · sj
The Hopfield network is a recurrent neural network that stores binary patterns and retrieves corrupted inputs through asynchronous updates. Each update monotonically decreases the energy function, guaranteeing convergence to a local minimum. Its storage capacity is roughly 0.138N patterns.
— ✦ —