To propagate a new pattern through an ART1 network without adjusting weights, i.e. to classify a pattern, two different update functions have been implemented:
Like the learning function, both of the update functions only take the vigilance value as parameter. It has to be entered in the remote panel, the line below the parameters for the learning function. The difference between the two update functions is the following:
ART1_Stable propagates a pattern until the network is stable, i.e. either the cl unit or the nc unit is active. To use this update function, you can use the -button of the remote panel. The next pattern is copied to the input units and propagated completely through the net, until a stable state is reached.
ART1_Synchronous, performs just one propagation step with each call. To use this function you have to press the -button to reset the net to a defined initial state, where each unit has its initial activation value. Then copy a new pattern into the input layer, using the buttons and . Now you can choose the desired number of propagation steps that should be performed, when pressing the -button (default is 1). With this update function it is very easy to observe how the ART1 learning algorithm does its job.
So use ART1_Synchronous, to trace a pattern through a network, ART1_Stable to propagate the pattern until a stable state is reached.