JARINGAN PEMBELAJARAN TIRUAN YANG DITURUNKAN DARI SISTEM SYARAF BIOLOGIS DENGAN MENGGUNAKAN MODEL HOPFIELD (BAGIAN-1)

Abstract

An artificial neural network is described that employs novel neuronal elements based on some recently revealed fundamental properties of biological neuronal networks. The dynamically stable associative learning (Dystal) network learns both correlations and anticorrelations by associating patterns through local interactions manifest only at the input of neuronal elements. The network can be configured to either classify or restore patterns simply by changing the number of output units. Dystal exhibits some desireable properties : performance of the network is stable with respect to network parameters over wide ranges of their values and over the size of the input fields. Neither global nor global feedback connections are required during learning. So that, the network is particularly suitable for hardware implementation. The training pattern may be noisy and need not be orthogonal. A very large number of pattern can be stored, network architecture is not restricted to multi-layer feed-forward or any other specific structure. For a known set of input patterns, the network weights can be computed a priori, in closed form, and computational effort scales linearly with the number of conections. These properties are described by Hopfield Neural Network.