Balancing Privacy-Utility of Differential Privacy Mechanism: A Collaborative Perspective

Table 3

Comparative analysis of privacy games and collaborative differential privacy.

	Mechanisms	Models	Players		Strategies		Privacy	Utility	Equilibrium	Properties
Stackelberg game between machine learning and data obfuscation [23]	Gaussian mechanism	Stackelberg game	User	Learner	Perturbing data independently	Proactively perturbing data	Privacy budget	Empirical risk minimization	Stackelberg equilibrium	Dynamic game with complete information
Private and truthful aggregative game [24]	Exponential mechanism	Aggregative game	Spectrum users	Mediator	Mixed strategy	Computing mixed strategy and suggesting mixed strategy to each user	Joint differential privacy	Incentive compatibility	Nash equilibrium	Mediated game with complete information
Privacy games [25]	Randomized response	Bayesian game	Selfish agent	Adversary	Randomized secret	Discover secret type	Privacy violation	Reward from getting the coupon	Bayesian Nash equilibrium	Static game with incomplete information
Privacy-preserving Stackelberg mechanism [26]	Laplace mechanism	Stackelberg game	Leader	Follower	Obfuscating sensitive data	Optimizing decisions while anticipating the reaction of the follower	Estimated coordination variables	Follower’s fidelity recovery	Complying differential privacy and ensuring outcomes of the privacy-preserving Stackelberg mechanism being close-to-optimality	Dynamic game with complete information
cDP	(Discrete) Laplace mechanism	Collaborative model	Data curator	Data analyst	Randomized perturb data	Statistical analysis	Required privacy budget	Approximate data utility	Privacy-utility balance	Dynamic game with incomplete information