.png){kind=icon}
.png){kind=icon}
Return
SilentNoise: Non-Interactive Noise Generation for Differential Privacy With Malicious Security
.png){kind=icon}
PRE
AI
R
T
K
N
DOI：10.1109/tdsc.2026.3694826.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Differential privacy (DP) is a standard notion of privacy-preserving data analysis. Traditionally, DP has been studied in the central model, where a trusted curator collects all private inputs and releases aggregation results preserving DP. However, this model creates a single point of failure, as all private data are held by a single party. We aim to decentralize additive noise DP mechanisms by leveraging secure multiparty computation (MPC) techniques. The main technical challenge is how to obliviously sample secret noise from specific probability distributions in the presence of malicious parties. Most existing maliciously secure protocols require multiple rounds of interaction, resulting in high communication and computation overheads. Moreover, the only known maliciously secure, non-interactive protocol suffers from a utility loss that grows exponentially with the number of computing parties. In this work, we propose <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SilentNoise</small>, a general framework for maliciously secure, non-interactive protocols for sampling DP noise, and present several instantiations that improve utility and efficiency both theoretically and empirically. In our protocols, any incorrect noise contributed by malicious parties is rejected and meaningful utility is guaranteed whenever the protocols succeed. After a one-time interactive setup that can be amortized over multiple uses, parties can non-interactively generate an unbounded number of shares of noise. This leads to a significant improvement in efficiency compared to the previous interactive protocols. Furthermore, our protocols improve the utility loss of the prior maliciously secure, non-interactive protocol by an exponential factor in the number of computing parties. They can even achieve a constant utility loss that is comparable to central-model DP mechanisms, under a strong honest-majority setting.
Keywords:
Secure multiparty computation
differential privacy
Journal
IF:
7.5
Papers:
2.4K
Citations:
9.6K
Organization
Cited Papers