A Tensor Compiler with Automatic Data Packing for Simple and Efficient Fully Homomorphic Encryption

Explore a groundbreaking 20-minute conference talk from PLDI 2024 that introduces Fhelipe, an innovative tensor compiler designed to simplify and enhance Fully Homomorphic Encryption (FHE) programming. Learn how this compiler addresses FHE's major challenges of high performance overheads and programming complexity by offering a user-friendly numpy-style tensor programming interface. Discover the key contribution of automatic data packing, which optimizes data layouts for tensors and efficiently packs them into ciphertexts. Understand how Fhelipe matches or exceeds the performance of hand-optimized FHE applications, particularly in deep neural networks, while significantly reducing code size and complexity. Gain insights into the compiler's evaluation on both state-of-the-art FHE accelerators and CPUs, and its potential to revolutionize secure computation on encrypted data.

[PLDI24] A Tensor Compiler with Automatic Data Packing for Simple and Efficient Fully Homomorphic(…)