An extremely diluted asymmetric network with graded response neurons
Stephan Mertens · Journal of Physics A Mathematical and General · 1991
The properties of an extremely diluted asymmetric network of neurons with a sigmoidal response function are investigated. It is shown that in the absence of thermal noise the storage capacity increases with decreasing gain of the response function. The information capacity of the network in the case of multistate patterns is discussed. It turns out that the network can process only a finite amount of information in spite of the unbounded amount of information that can be encoded in multistate patterns.