Algorithm 2: Obtaining neural network-based features from projected data on the PCs

$\mathrm{The} \mathrm{feature} \mathrm{matrix} \mathit{F} = \left(f_1 f_2 \dots f_K\right)$ $\mathrm{S}1: \mathrm{Build} \mathrm{an} \mathrm{MLP} \mathrm{network} \mathrm{with} \mathrm{one} \mathrm{hidden} \mathrm{layer} \mathrm{and} P$ hidden nodes (neuron). $\mathrm{S}2: \mathrm{Start} \mathrm{training} \mathrm{the} \mathrm{network} \mathrm{for} \mathrm{classifying} \mathrm{the} \mathrm{examples} \mathrm{represented} \mathrm{by} \mathrm{the} \mathrm{rows} \mathrm{of} \mathit{F}$. S3: Halt training in early iterations. S4: Calculate the outputs of the hidden layer $$h_i = sigmoid\left(\mathit{F}\mathit{W}+b\right)$$ $\mathrm{where}, \mathit{W}$$\mathrm{is} \mathrm{the} \mathrm{weight} \mathrm{matrix} \mathrm{between} \mathrm{input} \mathrm{and} \mathrm{hidden} \mathrm{layers} \mathrm{and} i = 1, \dots , P$ $\mathrm{S}5: \mathrm{Construct} \mathrm{the} \mathrm{hidden} \mathrm{layer} \mathrm{output} \mathrm{matrix} \mathit{H} = \left(h_1 h_2 \dots h_P\right)$$\mathrm{whose} \mathrm{size} \mathrm{is} NxP$. Although the algorithm ends in this step, the following step demonstrates the effectiveness of the generated features. $\mathrm{S}6: \mathrm{Train} \mathrm{a} \mathrm{classifier} (\mathrm{such} \mathrm{as} \mathrm{SVM} \mathrm{or} \mathrm{MDC}) \mathrm{by} \mathrm{the} \mathrm{rows} \mathrm{of} \mathrm{the} \mathrm{matrix} \mathit{H}$.