Table 1 Notations and definitions used in the proposed CyberDetect-MLP framework.
Symbol | Description |
|---|---|
\(\:X\) | Original feature value |
\(\:{X}^{{\prime\:}}\) | Normalized feature value after Min-Max scaling |
\(\:{X}_{min}\) | Minimum value of a feature |
\(\:{X}_{max}\)​ | Maximum value of a feature |
\(\:{h}^{\left(l\right)}\) | Output of the \(\:{l}^{th}\) hidden layer |
\(\:{h}^{\left(l-1\right)}\) | Input to the \(\:{l}^{th}\) hidden layer |
\(\:{W}^{\left(l\right)}\) | Weight matrix of the \(\:{l}^{th}\) layer |
\(\:{b}^{\left(l\right)}\) | Bias vector of the \(\:{l}^{th}\) layer |
\(\:\sigma\:\left(\bullet\:\right)\) | Activation function (ReLU) |
\(\:{z}_{j}\) | Logit output for class j before softmax activation |
\(\:K\) | Total number of classes (attack types) |
\(\:P\left(y=j\:|\:x\right)\) | Predicted probability of class j for input \(\:x\) |
\(\:{y}_{ij}\) | True label indicator for instance i and class j |
\(\:{\widehat{y}}_{ij}\) | Predicted probability for instance iii and class j |
\(\:{\theta\:}_{t}\) | Model parameter at iteration t |
\(\:\alpha\:\) | Learning rate |
\(\:{\widehat{m}}_{t}\) | Bias-corrected first moment estimate (mean of gradients) |
\(\:{\widehat{v}}_{t}\) | Bias-corrected second moment estimate (uncentered variance) |
\(\:\epsilon\:\) | Small constant to avoid division by zero in optimization updates |
\(\:{\alpha\:}_{t}\) | Learning rate at epoch t in cosine annealing schedule |
\(\:{\alpha\:}_{min}\)​ | Minimum learning rate |
\(\:{\alpha\:}_{max}\) | Maximum learning rate |
\(\:T\) | Total number of training epochs in the learning rate schedule |
