An autoassociative network in SNNS consists of two layers: A layer of world units and a layer of learning units. The representation on the world units indicates the information coming into the network from the outside world. The representation on the learning units represents the network's current interpretation of the incoming information. This interpretation is determined partially by the input and partially by the network's prior learning.
Figure 
shows a simple example network. Each unit in the
world layer sends input to exactly one unit in the learning layer.
The connected pair of units corresponds to a single node in the
typical representation of autoassociative networks. The link from the
world unit always has a weight of 1.0, and is unidirectional from the
world unit to the learning unit. The learning units are fully
interconnected with each other.
Figure: A simple Autoassociative Memory Network
The links between the learning units change according to the selected learning rule to fit the representation on the world units. The links between the world units and their corresponding learning units are not affected by learning.