Hopfield Networks

• Another type of net was developed by John Hopfield and is thus called the Hopfield Net.
• It can be used as a content addressable memory.
• Each neuron has a connection to all the other neurons.
• Each neuron has a firing threshold associated with.
• Both thresholds and connections can be negative.
• Running the system in synchronous mode (all neurons at once), if a neuron is on, it sends its weights to all of the neurons it is connected to.
• The next series of on's is calculated from this.
• A stable state is one where the same state repeats. (Once it repeats it will always repeat)
• Example:
• Widrow-Hoff learning rule and energy
• It's content addressable and robust, because you don't have to present the full pattern to get to a stable state.