next up previous contents
Next: Network Operation Up: ART Previous: Introduction

Network Characteristics

Both the ART architectures are two-layer networks in which units in the first, or input, layer receive inputs of the data and feedback from the second layer. Units in the second, or output, layer receive inputs from the input layer and interact among themselves.

The ART networks implement a nearest neighbour classification algorithm whose simplicity is obscured by the complexity of the network architecture.

The network stores patterns as sets of weights assigned to the paths connecting the input units to each of the output units. Presentation of an input vector causes the output units to be activated, the amount of activation depending on the similarity between the input vector and the stored pattern. Finding the nearest neighbour is simply a matter of deciding which output unit has been activated the most.

The networks also have a mechanism for adding new units to the output layer. This mechanism is invoked whenever an input is presented which is dissimilar to all the existing stored patterns. The invocation of this mechanism is regulated by a control parameter. In fact, the control of the operation of the network is a complex process requiring interactions between the units in both layers and a control unit.

The figure shows a simple ART1 network.


 
Figure 5.24: A simple ART1 network
\begin{figure} % latex2html id marker 4856 \setlength {\unitlength}{1mm} \beg... ...23,28){\line(0,-1){13}} \put(123,15){\vector(-1,0){18}}\end{picture}\end{figure}

next up previous contents
Next: Network Operation Up: ART Previous: Introduction
Mike Alder
9/19/1997