✅Verifiable FHE

Why "verifiable" so important for FHE

In the context of remote / delegated computation, FHE ensures privacy but NOT integrity. For instance, imagine a scenario where a computing node asserts it has generated an output using the LLAMA-70B model, yet in reality, it ran the computation with the LLAMA-7B model.

What are the challenges

ZK-SNARK provides natural solution for verifiability. However, challenge arises when integrating FHE and ZK-SNARK. The efficiency is far from satisfactory due to their different mathematical structures. FHE is built on ring structures while ZK-SNARKs runs on fields; this disparity results in poor performance.

Solution

We've been working on developing SPARC (Structure Preserved ARguments of Computation). It is a high performance proof system optimized for scenarios where the computational structure has unique characteristics. It is designed to preserve the inherent structure of these computations, resulting in more efficient proof generation.

Refer to SPARC for more details.