41、多模态加密恶意流量检测与智能城市众包平台

多模态加密恶意流量检测与智能城市众包平台

1. 多模态加密恶意流量检测

1.1 MEMTD 训练算法

为了检测加密的恶意流量，提出了一种名为 MEMTD 的多模态深度学习方法。其训练算法如下：

Algorithm 1: Training MEMTD
Input: Sample sets
Si = {(si1, y1), (si2, y2), ..., (siN, yN)}, i ∈{1, 2, 3}, yn ∈{1, 2, ..., K}
where s1n, s2n, s3n and yn represent the xhpb, xpls, xpais and label of the
n-th sample respectively. The number of samples is N, and the number
of traffic categories is K.
Output: A trained instantiated MEMTD network, Net.
Require: ConvNet, GruNet, FusionNet - 1 and FusionNet - 2.
1 Net1 ←ConvNet
2 Net2 ←GruNet // i.e., GruNet - 1
3 Net3 ←GruNet // i.e., GruNet - 2
4 for i in {1, 2, 3} do
5
    Train(Neti, Si) // pretraining stage
6
    Neti ←Neti dropped MLP layer and softmax
7
    Freez