EURASIP Journal on Information Security

Table 1 Training parameters

From: Secure machine learning against adversarial samples at test time

Parameter	MNIST
α_NormalNoise	0.44
α_BIM	0.44
α_CW	0.54
α_DF	0.54
τ_NormalNoise	0.1
τ_BIM	0.1
τ_CW	0.0265
τ_DF	0.0265
β	\(\frac {\alpha }{2}+\frac {1-\alpha }{11}\)
γ	β
ρ₀	0.1
Batch size	60
k_max	3000
Steps	100

Back to article page