Privacy Amplification by Subsampling in Time Domain

Releasing temporal aggregate signals with enough privacy guarantees is still tricky despite their wide applications and impact on society. The main difficulty lies in their sensitivity which scales linearly with the signal length. We analyze that one can reduce the sensitivity by subsampling in the time domain under reasonable assumptions. Then, based on the analysis, we propose a differentially private algorithm that utilizes signal subsampling and filtering. We demonstrate the utility gain of our algorithm empirically with the real and synthetic signals.