Metric-Normalized Posterior Leakage (mPL): Attacker-Aligned Privacy for Joint Consumption

TL;DR

This paper introduces metric-normalized posterior leakage (mPL), a privacy measure aligned with attacker inference, and proposes a practical framework (AmPL) to control privacy leakage in joint ML consumption scenarios.

Contribution

It formalizes probabilistically bounded mPL (PBmPL) and operationalizes it through Adaptive mPL (AmPL), improving privacy guarantees under joint observation.

Findings

Neural adversaries can violate mPL under joint consumption despite mDP.

AmPL reduces the frequency of privacy violations in a word-embedding case study.

AmPL maintains low utility loss while providing certifiable privacy protection.

Abstract

Metric differential privacy (mDP) strengthens local differential privacy (LDP) by scaling noise to semantic distance, but many machine learning (ML) systems are consumed under joint observation, where model-agnostic, per-record guarantees can miss leakage from evidence aggregation. We introduce metric-normalized posterior leakage (mPL), an attacker-aligned, distance-calibrated measure of posterior-odds shift induced by releases, and show that for single or independent releases, uniformly bounding mPL is equivalent to mDP. Under joint observation, however, satisfying mDP may still leave mPL high because learned aggregators compound evidence across correlated items. To make control practical, we formalize probabilistically bounded mPL (PBmPL), which limits how often mPL may exceed a target budget, and we operationalize it via Adaptive mPL (AmPL), a trust-and-verify framework that perturbs, audits with a learned attacker, and adapts parameters (with optional Bayesian remapping) to balance privacy and utility. In a word-embedding case study, neural adversaries violate mPL under joint consumption despite per-record mDP perturbations, whereas AmPL substantially lowers the frequency of such violations with low utility loss, indicating PBmPL as a practical, certifiable protection for joint-consumption settings.