[Previous] [Next]

The Storage of Stable Patterns

• The Hopfield network employs Hebbian learning. w_ij = a_i a_j
• If we instantiate a pattern on the array of units and apply the above rule that pattern becomes an attractor of the network.
• If the pattern of activation of the network is close to a stored pattern it will tend to migrate to that pattern.
• Once it arrives at the attractor it will remain there indefinitely - the attractors are stable.