AN ENERGY FUNCTION FOR SPECIALIZATION

We present a model of unsupervised learning based on the minimization of an energy function. The minima of the energy function are related to the degree of specialization of a certain class of artificial neuronal cells-grandmother cells-in the neural network model proposed by Haken. The self-organizing properties of the system are demonstrated by feeding input into a network of such cells with originally randomized synaptic connections. The relation of this learning algorithm to other learning schemes, like e.g. Kohonen's feature maps, is outlined. © 1990.