Website Fingerprinting Attacks Based on Homology Analysis

Table 1

Model hyperparameter search space and final selection.

Hyperparameters	Search space	Selected value
Number of filters
Conv2d1	[8 … 32]	16
Conv2d2	[16 … 64]	32
Normalization methods	[Batch Normalization, Layer Normalization]	Layer Normalization
Activation functions	[ReLU, ELU, LeakyReLU]	ReLU
Pooling layers	[Average, max]	Max
BiLSTM	[64 … 256]	128
Number of FC layers	[1 … 4]	3
[Filter, pool, stride] sizes	[2 … 8]	[3, 3, 1]
Loss function	[Cross-entropy loss]	Cross-entropy loss
Optimizer	[SGD, adam, Adamax, RMSProp]	Adam
Learning rate	[0.0001 … 0.01]	0.001
Training epochs	[10 … 50]	30
Minibatch size	[16 … 64]	48