Topology Learning Solved by Extended Objects: A Neural Network Model

We describe a self-organizing artificial neural network architecture that simultaneously creates (1) local filters with competitive learning, (2) a representation of the topology of the external world with Hebbian learning, and introduces (3) Kohonen type cooperative neighbour training through the self-developed connections. Such a network is capable of building up a 3-dimensional topology from two 2-dimensional images — similar to the working of the human eye — and it shows increased adaptivity.