MAE-CAD: An IP-Based Core Network Asset Discovery Technology Based on Multiple Autoencoders

Table 5

Structure composition of MAE-CAD with different numbers of AEs in the test set.

Pre-training	MAE	Softmax layer
No pretraining		32–2
Pretraining with 1 AE	AE 1 = (encoder: 40-1024-512-256-128-32, decoder:32-40)	32–2
Pretraining with 2 AEs	AE 1 = (encoder: 40-1024-512-256-128, decoder:128-40)	32–2
	AE 2 = (encoder: 128-32, decoder:32-128)
Pretraining with 3 AEs	AE 1 = (encoder: 40-1024-512, decoder:512-40)	32–2
	AE 2 = (encoder: 512-256-128, decoder:128-512)
	AE 3 = (encoder: 128-32, decoder:32-128)
Pretraining with 5 AEs	AE 1 = (encoder: 40-1024, decoder:1024-40)	32–2
	AE 2 = (encoder: 1024-512, decoder: 512-1024)
	AE 3 = (encoder: 512-256, decoder:256-512)
	AE 4 = (encoder: 256-128, decoder:128-256)
	AE 5 = (encoder: 128-32, decoder:32-128)