Authors:

Sreevickrant Sreekanth

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Fully Homomorphic Encryption (FHE) is a cryptographic technique that allows secure computations on encrypted data. However, FHE suffers from slow execution. Previous attempts to improve FHE performance through vectorization techniques have often overlooked the costly rotations of vector operands. To address this issue, we propose a novel approach that builds upon Coyote, a method known for effectively vectorizing computational kernels while minimizing rotations in encrypted circuits. Coyote addresses scheduling and data layout challenges by identifying vectorizable subcircuits with minimal data movement overhead. By conducting a joint search for optimal vectorization and lane placement, Coyote achieves efficient vector schedules and intelligent rotation schemes, resulting in significant speedups for computational kernels in FHE. Nevertheless, Coyote is hampered by long compilation times and excessive rotations when generating vector schedules for large circuits. To overcome these limitations, we leverage Coyote's capabilities by vectorizing smaller replicated subcircuits and then combining and interleaving them to generate a more efficient vector schedule for the large circuit. This approach not only results in faster compilation times but also reduces the number of excessive rotations. In summary, our project presents a novel approach to optimize FHE through vectorization by addressing the challenges of identifying repeated subcircuits that can be vectorized together.

Source:

Purdue University / 2023

Topics:

No topics listed

Co-authors:

Sreevickrant Sreekanth