80 likes | 173 Views
Privacy Meets Game Theory and Economics. Example: Pricing of an O nline Cartoon. (digital good, unlimited supply). Maximizing Revenue. What if I bid 0.5 instead?. 0.4. t=. 1. 0.3. 0.7. [0,1]. t i. p. p. OPT. p. n. The Implementation Challenge. t’ 1. t 1. t 2. t 2. t 3.
E N D
Example: Pricing of an Online Cartoon • (digital good, unlimited supply)
Maximizing Revenue What if I bid 0.5 instead? 0.4 t= 1 0.3 0.7 [0,1] ti p p OPT p n
The Implementation Challenge t’1 t1 t2 t2 t3 t3 t’= t= tn-1 tn-1 tn tn Player i does not Prefer M(t’) over M(t) Unilateral Deviation by player i M M M(t) M(t’)
b1 b’1 b2 b2 b3 b3 b’= b= bn-1 bn-1 bn bn distributions are ε-close Neighboring:One entrymodified A A The Differential Privacy challengeDwork, McSherry, Nissim, Smith A(b) A(b’)
What we Know • [NST]: Implementation in dominant strategies • Revenue recovered at least OPT – O(n2/3) • Generic construction, many potential applications • Ingredients: • Mechanism preserving differential privacy • Making agents (almost) indifferent to what strategy they use • Efficient (i.e., recovers most of OPT) • “Punishing” mechanism • Making agents truthful • Inefficient
What about you? • Learn about differential privacy and mechanism design • Extend known results • E.g., new applications • Examine notions of privacy • Explore relationships between privacy and game-theoretic concepts • Using what? • Your knowledge in probability/statistics, algorithms, …, mathematics
Further Details … • Contact me • kobbi@cs.bgu.ac.il