Private ways to estimate high-dimensional functions

Researchers designed evaluators to privately approximate function values using a black-box approach.

Researchers studied the private black-box evaluation of arbitrary high-dimensional functions. The goal is to privately output an approximation to f(S)f(S) given a dataset SS and query access to a function ff.

The sensitivity of the function ff may be unknown or unbounded. A black-box evaluator can only query the value of ff over a down neighborhood of SS. A down neighborhood consists of subsets of SS obtained by removing a bounded number of elements.

The researchers designed generic black-box evaluators for both pure and approximate differential privacy. These evaluators use a reduction to high-dimensional private quasi-concave optimization (QCO). Quasi-concave optimization (QCO) is a type of optimization.

The evaluator's output is guaranteed to be in the convex hull of vectors obtained by applying ff on sets in a down neighborhood of SS with high probability. This work generalizes the result of Linder et al. [LRSS25] for real-valued function evaluation to high dimensions.

The evaluators can be applied to user-level quasi-concave and convex optimization. They may improve upon the sample-and-aggregate baseline of Nissim et al. [NRS07] by replacing multiplicative data overhead with an additive overhead.

The formulation of high dimensional black-box evaluation and high dimensional private QCO are equivalent up to a factor of the dimension. The hardest private evaluation instances arise when ff is a non-private quasi-concave optimizer.

Key facts

  • The evaluator outputs an approximation to f(S)f(S) privately.
  • The evaluator uses a down neighborhood of SS to query the function ff.
  • The work generalizes the result of Linder et al. [LRSS25] to high dimensions.
  • The evaluators may improve upon the sample-and-aggregate baseline of Nissim et al. [NRS07].

All sections

Edition