Learning versus Refutation in Noninteractive Local Differential Privacy
Edmonds, Alexander, Nikolov, Aleksandar, Pitassi, Toniann
–arXiv.org Artificial Intelligence
We study two basic statistical tasks in non-interactive local differential privacy (LDP): learning and refutation. Learning requires finding a concept that best fits an unknown target function (from labelled samples drawn from a distribution), whereas refutation requires distinguishing between data distributions that are well-correlated with some concept in the class, versus distributions where the labels are random. Our main result is a complete characterization of the sample complexity of agnostic PAC learning for non-interactive LDP protocols. We show that the optimal sample complexity for any concept class is captured by the approximate $\gamma_2$~norm of a natural matrix associated with the class. Combined with previous work [Edmonds, Nikolov and Ullman, 2019] this gives an equivalence between learning and refutation in the agnostic setting.
arXiv.org Artificial Intelligence
Oct-25-2022
- Country:
- North America
- United States
- Massachusetts > Middlesex County
- Cambridge (0.04)
- Georgia > Fulton County
- Atlanta (0.04)
- Massachusetts > Middlesex County
- Canada > Ontario
- Toronto (0.14)
- United States
- Europe > Netherlands
- North Holland > Amsterdam (0.04)
- North America
- Genre:
- Research Report (0.64)
- Industry:
- Information Technology > Security & Privacy (0.67)
- Technology: