ε varepsilon ε-differential privacy的(利用Max Divergence)等价定义 : quad A randomized mechanism f f f is ε varepsilon ε-differentially private if and only if its distribution over any two adjacent inputs D D D and D ′ D' D′ satisfies : D ∞ ( f ( D ) ∥ f ( D ′ ) ) ≤ ε ,, D_infty big( f(D) | f(D') big) le varepsilon D∞(f(D)∥f(D′))≤ε quad 注: 此种等价定义在《The Algorithmic Foundations of Differential Privacy》一书中也有提及,详见 P 44 mathcal P_{mathcal{44}} P44
在此基础上进行拓展,(利用Renyi Divergence)得到Renyi differential privacy : quad A randomized mechanism f f f: D ↦ R mathcal D mapsto mathcal R D↦R is said to have ε varepsilon ε-Renyi differential privacy of order α alpha α, or ( α alpha α, ε varepsilon ε)-RDP for short, if for any adjacent D D D, D ′ D' D′ ∈ D in mathcal D ∈D it holds that : D α ( f ( D ) ∥ f ( D ′ ) ) ≤ ε ,, D_alpha big( f(D) | f(D') big) le varepsilon Dα(f(D)∥f(D′))≤ε quad 注: Renyi Divergence的定义中令 α alpha α趋于无穷即得到 Max Divergence