A Privacy-Preserving and Edge-Collaborating Architecture for Personalized Mobility

Table 2

The hyperparameters.

Variables	Description	Value
	Sample size of the PID under FPMS	According to the evaluation setting
	Sample size of the PRC under CPMS	According to the evaluation setting
	Training workload per sample	10
	Computation coefficient	1
	Bandwidth	1 MHz
	Communication coefficient	50
	Transmit power
	Power spectral density of the Gaussian noise
	A unitless constant	−31.54
	Model size	1 kb
	Data size per sample	7 kb
	CPU frequency of the PRC	0.5–2.5 GHz
	CPU frequency of the PID	3 GHz
	Path loss exponent	3
	Distance to server	0–500 m
	Covariance of channel shadowing	3.65
	Maximum delay	0.5 ms