In Differential Privacy, There is Truth: On Vote Leakage in Ensemble Private Learning

TL;DR

This paper reveals that noise added in differentially private ensemble learning methods like PATE can inadvertently enable sensitive information leakage through vote histograms, challenging assumptions about privacy guarantees.

Contribution

It uncovers a new form of privacy leakage in differentially private ensemble models by exploiting stochastic vote distributions, highlighting limitations of current privacy guarantees.

Findings

Adversaries can extract sensitive attributes from vote histograms.

Adding more noise can paradoxically make the attack easier.

The attack does not violate differential privacy but breaches privacy norms.

Abstract

When learning from sensitive data, care must be taken to ensure that training algorithms address privacy concerns. The canonical Private Aggregation of Teacher Ensembles, or PATE, computes output labels by aggregating the predictions of a (possibly distributed) collection of teacher models via a voting mechanism. The mechanism adds noise to attain a differential privacy guarantee with respect to the teachers' training data. In this work, we observe that this use of noise, which makes PATE predictions stochastic, enables new forms of leakage of sensitive information. For a given input, our adversary exploits this stochasticity to extract high-fidelity histograms of the votes submitted by the underlying teachers. From these histograms, the adversary can learn sensitive attributes of the input such as race, gender, or age. Although this attack does not directly violate the differential privacy guarantee, it clearly violates privacy norms and expectations, and would not be possible at all without the noise inserted to obtain differential privacy. In fact, counter-intuitively, the attack becomes easier as we add more noise to provide stronger differential privacy. We hope this encourages future work to consider privacy holistically rather than treat differential privacy as a panacea.