Discrete Gaussian Sampling for Low-Power Devices

Thesis (Master's)--University of Washington, 2015

Sampling from the discrete Gaussian probability distribution is used in lattice-based cryptosystems. A need for faster and memory-efficient samplers has become a necessity for improving the performance of such cryptosystems. We propose a new algorithm for sampling from the Gaussian distribution that can efficiently change on-the-fly its speed/memory requirement. The Ziggurat algorithm that attempted to do this requires up to 1000 seconds of computation time to change memory requirements on-the-fly. Our algorithm eliminates this large computational overhead.