The best ANN architecture had 18 input neurons, 43 hidden neurons and 1 output neuron, trained using the Levenberg Marquardt learning algorithm using tangent sigmoid equation as the activation and output functions.
To quantify the effects of the bias (Fig. 5a), we summed the vectors corresponding to the preferred direction of each neuron trained under a given bias.
A central composite design was applied to find optimum values of number of neurons, training epoch, step size, training percentage and momentum coefficient.
Finally, the optimum values of variables to obtain minimum response were 22, 7670, 0.28, 65% and 0.85 for number of neurons, training epoch, step size, training percentage and momentum coefficient, respectively.
Table 3 Performance comparisons for the IS and VC problems Data sets Method N H neurons Training time(s) Testing η o Testing η a IS SVM 96a 11.61 90.62 – MRAN 78 11.68 85.82 – GAP-RBF 87 5.77 86.34.
Computational experiments were designed to cross-examine the two types of hidden layers of networks with different hidden neurons, training tolerances, and testing tolerances based on the v-fold cross-validation technique.
A subset of neurons, called training neurons, intermittently receive superthreshold external input.
This is done by selecting a subset of neurons, called training neurons (TN), that receive strong excitatory external input, inducing them to spike synchronously at selected times.
The main differences between the various types of ANNs include network architecture consisting of hidden layer number, the number of neurons in each layer and activation function used in each neuron and training methods.
Unlike blocking octopamine neurons, blocking the rewarding dopamine neurons during training also impaired nutritious sucrose-conditioned appetitive memory [56,90].
