Secret Noise Table Sharing for Low-Overhead Differential Privacy

The patent segments the probability distribution approximation into multiple smaller tables, each handling a specific portion of the distribution. Instead of using one huge table, the system divides the approximation task across multiple terminals, where each terminal stores and processes a segment of the overall probability distribution. This segmentation reduces the memory burden on individual terminals and distributes communication traffic across the network.

The patent implements a nested structure where multiple layers of tables are used to approximate the probability distribution. Each table is nested within a broader probabilistic framework, with coarser approximations at outer layers and finer details at inner layers. This nested approach allows accurate probability distribution approximation while keeping individual table sizes manageable, thereby reducing memory usage and communication overhead.

The patent performs preliminary actions by pre-computing and storing probability distribution tables in encrypted form before the actual noise generation process. These pre-computed tables are prepared in advance and stored securely, so that during runtime, the system can directly retrieve and use them without performing complex computations. This preliminary preparation maintains security through encrypted storage while significantly improving noise generation speed during actual use.

The patent uses multiple copies of the same probability distribution table distributed across different terminals. Instead of one terminal performing all secure computations, the system creates replicated copies of the encrypted table and distributes them to multiple participants. This copying approach allows parallel processing and reduces the computational burden on any single terminal, thereby improving overall noise generation productivity while maintaining security through distributed secure computation.