The weights of the links from F 1 to F 2 are a little more di cult to explain. The F 2 ! F 1 links are all set to 1.0 as described in. These links are the ones from F 1 to F 2 and the ones from F 2 to F 1 respectively. To use an ART1 network in SNNS several functions have been implemented: one to initialize the network, one to train it and two di erent update functions to propagate an input pattern through the net.ĪRT1 Initialization Function First the ART1 initialization function ART1 Weights has to be selected from the list of initialization functions.ĪRT1 Weights is responsible to set the initial values of the trainable links in an ART1 network. The gain units g 1 and g 2 with their known functions and at last the units ri (reset input), rc (reset comparison), rg (reset general) and (vigilance), which realize the reset function.įor an exact de nition of the required topology for ART1 networks in SNNS see section 9.13.4 The nc unit indicates an unclassi able pattern, when active. The cl unit gets positive activation when the input pattern has been successfully classi ed. Finally, there are several special units.

Rst capa cl manual