FD Fengnan Deng

Research Project

Private Minimum Hellinger Distance Estimation via Hellinger Distance Differential Privacy

A privacy framework that unifies and extends differential privacy notions while keeping estimation both robust and statistically efficient.

arXiv 2025 Differential Privacy Robust Inference

Overview

  • Proposes Hellinger Differential Privacy as a privacy framework that unifies and extends existing notions of differential privacy.
  • Develops private minimum Hellinger distance estimators that target robustness and statistical efficiency at the same time.
  • Designs private gradient descent and Newton-Raphson procedures for practical computation.

Methods and contribution

This work asks how privacy guarantees can be built into estimation without discarding the robustness benefits of Hellinger-based inference. The result is a new privacy lens that supports both theoretical guarantees and implementable estimators.

Numerical experiments in Matlab compare the proposed framework with other privacy mechanisms, highlighting how private optimization and robust divergence-based estimation can work together in sensitive data settings.