In his seminal paper Cover used geometrical arguments to compute the probability of separating two sets of patterns with a perceptron. We extend these ideas to feedforward networks with hidden layers. There are intrinsic limitations to the number of patterns that a net of this kind can separate and we find quantitative bounds valid far any net with d input and h hidden neurons.
Properties of Feed-Forward Neural Networks
BUDINICH, MARCO;MILOTTI, EDOARDO
1992-01-01
Abstract
In his seminal paper Cover used geometrical arguments to compute the probability of separating two sets of patterns with a perceptron. We extend these ideas to feedforward networks with hidden layers. There are intrinsic limitations to the number of patterns that a net of this kind can separate and we find quantitative bounds valid far any net with d input and h hidden neurons.File in questo prodotto:
Non ci sono file associati a questo prodotto.
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
