摘要 |
<p>An artificial neural network, for example a Hopfield network, includes a number of interconnected neurons, one of which (4) is shown individually and the others are shown schematically as a weight matrix (1). A bi-level threshold circuit (5) receives a weighted sum of the outputs of other neurons from the matrix (1) and a noise signal from a noise source (n(t)). The output from the threshold circuit is filtered by a low pass filter (2) and the resulting activation function for the neuron approximates to a continuously variable non-linear function. <IMAGE></p> |