The connectivity between two layers is bidirectional and not unidirectional as in the previous system . This is necessary for two reasons: Firstly, by this means the running blobs of the two connected layers can more easily align. With unidirectional connections one blob would systematically run behind the other. Secondly, connections in both directions are necessary for a recognition system. The connections from model to image layer are necessary to allow the models to move the attention blob in the image into a region that fits the models well. The connections from the image to the model layers are necessary to provide a discrimination cue as to which model best fits the image. Otherwise, each model would exhibit the same level of activity.