The Discrete Gaussian Expectation Maximization (Gradient) Algorithm for Differential Privacy

Figure 2

Estimation error of GMM w.r.t. privacy budget , data dimension (lower) , data size , and iteration . (a) n = 2000; d = 10, (b) n = 2000;  = 0.5, and (c) d = 10,  = 0.5.

(a)

(b)

(c)