English

Private Realizable-to-Agnostic Transformation with Near-Optimal Sample Complexity

Machine Learning 2025-10-03 v1 Machine Learning

Abstract

The realizable-to-agnostic transformation (Beimel et al., 2015; Alon et al., 2020) provides a general mechanism to convert a private learner in the realizable setting (where the examples are labeled by some function in the concept class) to a private learner in the agnostic setting (where no assumptions are imposed on the data). Specifically, for any concept class C\mathcal{C} and error parameter α\alpha, a private realizable learner for C\mathcal{C} can be transformed into a private agnostic learner while only increasing the sample complexity by O~(VC(C)/α2)\widetilde{O}(\mathrm{VC}(\mathcal{C})/\alpha^2), which is essentially tight assuming a constant privacy parameter ε=Θ(1)\varepsilon = \Theta(1). However, when ε\varepsilon can be arbitrary, one has to apply the standard privacy-amplification-by-subsampling technique (Kasiviswanathan et al., 2011), resulting in a suboptimal extra sample complexity of O~(VC(C)/α2ε)\widetilde{O}(\mathrm{VC}(\mathcal{C})/\alpha^2\varepsilon) that involves a 1/ε1/\varepsilon factor. In this work, we give an improved construction that eliminates the dependence on ε\varepsilon, thereby achieving a near-optimal extra sample complexity of O~(VC(C)/α2)\widetilde{O}(\mathrm{VC}(\mathcal{C})/\alpha^2) for any ε1\varepsilon\le 1. Moreover, our result reveals that in private agnostic learning, the privacy cost is only significant for the realizable part. We also leverage our technique to obtain a nearly tight sample complexity bound for the private prediction problem, resolving an open question posed by Dwork and Feldman (2018) and Dagan and Feldman (2020).

Cite

@article{arxiv.2510.01291,
  title  = {Private Realizable-to-Agnostic Transformation with Near-Optimal Sample Complexity},
  author = {Bo Li and Wei Wang and Peng Ye},
  journal= {arXiv preprint arXiv:2510.01291},
  year   = {2025}
}
R2 v1 2026-07-01T06:11:34.341Z