Currently, we are exploring the use of another type of machine learning method-grammatical evolution (GE) – to evolve the inputs, weights, and architecture of NNs [ 60].
W neural network input weight.
(c) The input weight matrix W in is randomly created according to a uniform distribution.
a ki, c ki, b ki are, respectively, the input weight, bias term, and output weight of the i th neuron in the kth block.
In our RMCVELM system, the input weight w i is randomly generated between −0.5 and + 0.5, and b i is randomly generated between 0 and + 1.
By definition, d=d 1+d 2+d R. The input weight and the pattern can be computed via heuristic searching of the trellis of G.
where WSAE is the input weight coefficient of each image block connecting the SAE hidden layer with the whitening processing, b1 represents the input bias, and the σ is the activation function.
Similarly, Eq. 10 shows that strengthening a connection occurs by adding a value proportional to the input weight and in the form of an exponential function of the weight itself.
The proposed method automatically adjusts the prediction horizon P, the diagonal elements of the input weight matrix Λ, and the diagonal elements of the output weight matrix Γ for the sake of good performance.
Table 1 Input weight and output weight distribution at d=F=24 of compound code G in (4), g= [ 23, 35, 37] w 1 w 2 d 1 d 2 d R 0 12 0 12 12 12 0 12 0 12 12 12 12 12 0. Since the BER is linearly proportional to the UPEP (overline {text {PEP}}^{text {NC}}left (d|mathbf {W}_{d}right)) via corresponding input weights, the diversity order of NCC is equal to diversity order of the UPEP.
