Search papers, labs, and topics across Lattice.
This paper benchmarks secure multi-party computation (SMPC) and fully homomorphic encryption (FHE) across various ML tasks, providing a practical guide for practitioners. The authors theoretically compare SMPC and FHE, then empirically evaluate open-source software implementations for each. Benchmarking reveals that FHE outperforms SMPC for regressions and simple dense networks (especially with GPUs), while SMPC is faster for complex CNN models.
Choosing between secure multi-party computation (SMPC) and fully homomorphic encryption (FHE) for secure ML depends heavily on the model architecture: FHE excels at regressions and simple networks, while SMPC dominates for complex CNNs.
As security demands increase, the importance of secure computation technologies grows, yet these technologies can often seem overwhelming to practitioners. Furthermore, many approaches focus only on a single technology, potentially overlooking superior alternatives. This work aims to address the issue of selecting the right technology for secure computation by presenting a comparative analysis of two highly relevant cryptographic methods and their software implementations, with a particular focus on machine learning. Firstly, we provide a theoretical summary and comparison of the secure computation paradigms of secure multi-party computation (SMPC) and fully homomorphic encryption (FHE). We outline the advantages and limitations of the protocols, as well as the relevant open-source software implementations. Secondly, we present the results of extensive benchmarking of the main software frameworks identified for machine learning operations and models. Regarding the current state of the art in FHE, we observe that it outperforms SMPC for regressions. Additionally it may be faster for simple dense networks using GPUs or Hybrid Models. Conversely, SMPC showed superior performance for complex models such as CNNs. Our results should pave the way for more technology-agnostic benchmarking of secure computation technologies for machine learning, providing guidance for practitioners looking to adopt these technologies.
