Artificial Intelligence and Soft Computing: Behavioral ... - Arteimi.info

14.2 Artificial Neural Nets
An artificial neural net is an electrical analogue of a biological neural net [23].
The cell body in an artificial neural net is modeled by a linear activation
function. The activation function, in general, attempts to enhance the signal
contribution received through different dendrons. The action is assumed to be
signal conduction through resistive devices. The synapse in the artificial neural
net is modeled by a non-linear inhibiting function, for limiting the amplitude
of the signal processed at cell body.
Out
Out
Net
Out = 1/(1+e –Net )
Out
Net
Out = tanh (Net/2)
(a) Sigmoid function (b) tanh function
Net
Out = +1, Net > 0
= -1, Net < 0
= undefined, Net =0.
Out
Net
Out = 1, Net > 0
= 0, Net = 0
= undefined, Net < 0.
(c) Signum function (d) Step function
Fig.14.2: Common non-linear functions used for synaptic inhibition. Soft nonlinearity:
(a) Sigmoid and (b) tanh; Hard non-linearity: (c) Signum and
(d) Step.