In Layer two the cosine similarity is calculated for the node which is connected to the previous layer. Then the similarity scores are calculated for nodes which can be connected, and when it finds the nearby very best, It moves to the following Layer. this may transpire for all levels. Then Top