November 16, 2023
Fully Homomorphic Encryption (FHE) enables secure processing on encrypted data, maintaining privacy across a range of applications. However, the practical implementation of FHE has been constrained by substantial computational overhead. Recent research has explored vectorization strategies to optimize FHE computations, yet it often overlooks the problem of excessive vector operand rotations in encrypted circuits. This research talk leverages Coyote, an effective FHE compiler, capable of vectorizing computational kernels while minimizing rotations. Coyote, however, encounters challenges with large circuits which can lead to excessive rotations and lengthy compilation times. To address this challenge, we introduce PEAVS, an approach that breaks down extensive arithmetic circuits into smaller replicated subcircuits, vectorizes these subcircuits, and then composes them back into a schedule for the larger circuit. We demonstrate that this strategy significantly improves compilation times, reduces excessive rotations, and minimizes vector instructions, resulting in more efficient and scalable compilation for large circuits.
About Dulani Wijayarathne and Vickrant Sreekanth
Dulani is currently a senior undergraduate student majoring in Computer Engineering. Vickrant is a junior majoring in Computer Engineering.