Research‎ > ‎Distance Learning‎ > ‎

Hopfield Network Demo

In this demo we're going to train the Hopfield neural network with three patterns and then test its performance on a previously unseen testing pattern. The patterns are presented as square matrices that resemble letters of an alphabet (e.g. Σ, Π, Α). The three training patterns are depicted in the lower part of the figure while the testing input on the top of the figure. The units of the patterns have binary values (-1 or +1) that are depicted with red and blue dots respectively. By clicking on these dots you can switch the values of the training and testing patterns to your will. If you modify the training patterns you would have to readjust the weigths of the network (Hebbian learning) by pressing the button of the respective function. When you press the start button the demo will illustrate in green color the neuron of the network that is updated at each time step of the dynamics. The process is complete when the network converges to a stable state and a message box with the similarities before and after the dynamics is presented.