×
Not All Learnable Distribution Classes are Privately Learnable. We give an example of a class of distributions that is learnable in total variation distance with a finite number of samples, but not learnable under (\varepsilon, \delta)-differential privacy. This refutes a conjecture of Ashtiani.
Feb 1, 2024
Mar 15, 2024 · Abstract. We give an example of a class of distributions that is learnable in total variation distance with a finite number of samples, but not ...
Feb 1, 2024 · We give an example of a class of distributions that is learnable in total variation distance with a finite number of samples, but not learnable ...
Given samples from a distribution P belonging to some class of distributions E, can we output a distribution P1 that is close to P in total variation distance?
Not All Learnable Distribution Classes are Privately Learnable. Mark Bun, Gautam Kamath, Argyris Mouzakis, Vikrant Singhal. February 2024.
We consider the problem of online classification under a privacy constraint. In this setting a learner observes sequentially a stream of labelled examples ...
Jan 28, 2022 · In particular, it does not depend on |H|. Theorem 3 (Littlestone Classes are Privately Learnable). Let H ⊆ {±1}X be a class with Lit- tlestone ...
For any class H, under the realizability assumption, if there is a (0.1, 0.1, 0.1)-pure private learner for H, then H is privately learnable by a pure private ...
Aug 16, 2022 · The converse direction—that every DP-learnable class has a finite Littlestone dimension—utilizes an intimate relationship between thresholds and ...