Neural network model recalling spatial maps

Abstract

When driving through a place we have been before, we can recall and imagine the scenery that we shall see soon. Triggered by the newly recalled image, we can also recall other scenery further ahead of us. This paper offers a neural network model of such a recalling process. The model uses a correlation matrix memory for memorizing and recalling patterns. A correlation matrix memory by itself, however, does not accept shifts in location of stimulus patterns. In order to place stimulus patterns accurately at the location of one of the memorized patterns, we propose using the cross-correlation between the stimulus pattern and the "piled pattern". A map of Europe is divided into a number of overlapping segments, and these segments axe memorized in the proposed model. A map around Scotland is input to the model as the initial image. Triggered by the initial image, the model recalls maps of other parts of Europe sequentially up to Italy, for example.