Concepts de base
We provide efficient local Lipschitz filters for functions with bounded range, circumventing previous lower bounds. These filters enable applications to arbitrary real-valued functions, including private data analysis and tolerant testing.
Résumé
The paper studies local Lipschitz filters for real-valued functions with bounded range. Previous work only considered unbounded-range functions, and showed that local Lipschitz filters for such functions require exponential time in the dimension.
The authors demonstrate that important applications of local Lipschitz filters can be accomplished with filters for functions with bounded range. They provide two efficient local Lipschitz filters:
An ℓ1-respecting filter that runs in time (dr·polylog(n/δ))O(log(r/γ)) and outputs a (1+γ)-Lipschitz function g that is 2-close to the input function f in ℓ1-distance.
An ℓ0-respecting filter that runs in time dO(r)polylog(n) and outputs a 1-Lipschitz function g that is 2-close to f in ℓ0-distance.
The authors also prove a lower bound showing that the dependence on r in the running time of their filters is nearly optimal.
The authors showcase two applications of their local Lipschitz filters:
A differentially private mechanism for releasing outputs of arbitrary real-valued functions, even in the presence of malicious clients. The mechanism runs in time 2polylog min(r,nd) and has accuracy comparable to the Laplace mechanism for Lipschitz functions.
The first nontrivial tolerant tester for the Lipschitz property of functions f: {0,1}d → R, with query and time complexity 2^(eO(√d)).