Differential Privacy Principal Component Analysis for Support Vector Machines

Table 1

The notations used in this paper.

Notations	Meaning
X	A set of samples
	i-th piece of sample
n	The number of samples
d	The number of attributes
	The value space of samples
A	The covariance matrix of X
E	A noise matrix that obeys the Laplace distribution
	The covariance matrix after adding noise
	The i-th eigenvalue of the matrix
	The i-th eigenvector of the matrix
k	The number of principal components
	The principal component space, composed of eigenvectors corresponding to the first k eigenvalues
Y	The set of low-dimensional samples
	The normal vector of classification hyperplane
B	The intercept of classification hyperplane
C	The penalty factor
	The privacy budget
	The privacy parameter