In Differential Privacy, There is Truth: On Vote Leakage in Ensemble Private Learning
This work highlights a critical privacy vulnerability in a widely used differential privacy method, showing that formal guarantees can be undermined in practice, which is significant for researchers and practitioners in privacy-preserving machine learning.
The paper identifies that the noise added in the Private Aggregation of Teacher Ensembles (PATE) method for differential privacy enables adversaries to extract sensitive attributes like race, gender, or age from vote histograms, with the attack becoming easier as more noise is added for stronger privacy.
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.