Accelerating Multiparty Noise Generation Using Lookups

There is growing interest in combining Differential Privacy (DP) and Secure Multiparty Computation (MPC) to protect distributed database queries from both computational parties and those observing the result. This requires implementing both query evaluation and noise generation within MPC. While secure query evaluation is well-supported by existing MPC techniques, generating noise efficiently remains a challenge due to the nonlinearity of common sampling algorithms. We propose a new approach for multiparty noise sampling using recent advances in MPC lookup table (LUT) evaluations. Our method is distributionagnostic and maps a cheaply sampled index to a target noise distribution via oblivious LUT evaluation. We demonstrate the flexibility by approximating the discrete Laplace and Gaussian distributions to a negligible statistical distance. Our implementation, based on 3party replicated secret sharing (RSS), achieves sub-kilobyte communication and millisecondlevel computation. Per 1000 discrete Laplace or Gaussian samples, we require just 362 bytes of communication and under 1 ms per party (semi-honest setting). With recent batched multiplication checks, the amortized malicious setting adds less than 1 byte and 10 ms per sample. Our open-source implementation also extends MAESTRO-style LUT trade-offs, offering potential independent value.