Alberto Rossi (DIISM, University of Siena)

March 1, 2018 – 10:30 AM

DIISM, Artificial Intelligence laboratory (room 201), Siena SI

Description

Fully recursive perceptron network (FRPN) could be an alternative to normal MLP. It is composed by three layer, input, output and only one hidden with its neurons interconnected in instantaneous way.

FRPN computational capability exceeding that of MLP with multiple hidden layers. It is shown that FRPN with a single hidden layer can simulate MLP with arbitrary number of hidden layers and it require less neuron in the hidden layer than MLP to match the performance. Moreover it is interesting because there isn’t need of select the number of layers.

Resources

FRPN paper

FRPN presentation