Feed-forward networks composed by neurons with activation functions of different parity
Edgardo A. Ferrán, R P J Perrazzo · Journal of Physics A Mathematical and General · 1990
The authors study feed-forward networks of formal neurons having even activation functions. They show that networks of this kind have different computational properties than the ones with neurons having odd processing functions. They show that networks containing mixtures of this two types of neurons have richer representability properties. They extend their results to cases of discrete processing. These properties have been checked in numerical simulations performed in small enough systems to allow for an explicit enumeration of all synaptic matrices and Boolean functions.