OpenCBD: A Network-Encrypted Unknown Traffic Identification Scheme Based on Open-Set Recognition

Table 5

Experimental hyperparameters.

Hyperparameter name	Hyperparameter value
Epoch	80 in individual model, 60 in ensemble model
Batch	64
Plaintext packets in pretraining	256 bytes of plaintext
Ciphertext packets in pretraining	Randomly selected from ISCXVPN2016, excluding the 13 classes mentioned above, and classes and sizes are not fixed
Sample numbers of positive and negative sample sets in pretraining	5000, 5000
in individual model	3 or 4 or 5 or 6 or 7
Dimensions of CNN module input vectors	256
Number of convolution channels in CNN module	1
Convolution kernel size in CNN module	3
Convolution stride size in CNN module	1
Pooling size in CNN module	3
Dimensions of CNN module output vectors	242
Dimensions of fully connected layer outputs in the encoder	240
in Concat layer	15
Dimensions of BERT module input matrix
in BERT module	8
Head numbers of multihead attention in each transformer encoder layer	12
Dimensions of BERT module output vectors	240
Dimensions of fully connected layer outputs in classifier	in individual model, in ensemble model
Number of encoders in ensemble model	4 or 8 or 12
in ensemble model	8
in open-set testing