A neural network model for the developmental process of hypercomplex cells

A self-organizing neural network model is proposed which gives a possible implementation of the developmental process of hypercomplex cells in the mammalian visual cortex. It is composed of connections from complex cells to hypercomplex cells via an excitatory fixed synapse or an inhibitory modifiable synapse. Two types of receptive fields termed `single-stopped´ and `double-stopped´ are formed from the same network. The relation between values of parameters in the model and the type of receptive fields is made clear.