A. Sperduti et A. Starita, SUPERVISED NEURAL NETWORKS FOR THE CLASSIFICATION OF STRUCTURES, IEEE transactions on neural networks, 8(3), 1997, pp. 714-735
Until now neural networks have been used for classifying unstructured
patterns and sequences, However, standard neural networks and statisti
cal methods are usually believed to be inadequate when dealing with co
mplex structures because of their feature-based approach, In fact, fea
ture-based approaches usually fail to give satisfactory solutions beca
use of the sensitivity of the approach to the a priori selection of th
e features, and the incapacity to represent any specific information o
n the relationships among the components of the structures, However, w
e show that neural networks can, in fact, represent and classify struc
tured patterns, The key idea underpinning our approach is the use of t
he so called ''generalized recursive neuron,'' which is essentially a
generalization to structures of a recurrent neuron, By using generaliz
ed recursive neurons, all the supervised networks developed for the cl
assification of sequences, such as backpropagation through time networ
ks, realtime recurrent networks, simple recurrent networks, recurrent
cascade correlation networks, and neural trees can, on the whole, be g
eneralized to structures. The results obtained by some of the above ne
tworks (with generalized recursive neurons) on the classification of l
ogic terms are presented.