Table 4.

Comparison of DP-models.

Model	Similarity	Difference
DP-SGD	Add noise to gradient, they are DP-SGD algorithm, and it is variant	Add Gaussian noise to SGD Moments Accountant
The improvement of DP-SGD		Dynamic privacy budget allocation privacy accounting methods for different batch processing methods
Adaptive Laplace Mechanism		Adaptively add more noise to features that are not related to the model output
dPA	Based on functional mechanism, approximate functions to polynomial forms and perturb objective function mainly used for optimal algorithm	Uses Taylor expansion to approximate the cross-entropy error function to a polynomial form and add noise
PCDBN		Use Chebyshev polynomials to derive polynomial approximations of nonlinear objective functions and add noise
PATE		Decentralized training model adds noise to decisions and introduces student models