A new study presents a unified system-level characterization of secure multiparty computation (MPC) and fully homomorphic encryption (FHE) for privacy-preserving machine learning inference. The work evaluates two MPC variants—arithmetic/binary sharing conversion and function secret sharing—alongside FHE across multiple CNN and Transformer models.
- The evaluation covers a range of LAN and WAN environments, model sizes, batch sizes, and input sequence lengths.
- It assesses not only online inference latency but also energy consumption and monetary cost, accounting for offline and online computation and communication overheads.
- The analysis provides empirical guidance for selecting and optimizing these privacy-preserving compute paradigms.
- It outlines how evolving hardware and network trends are likely to shift trade-offs between MPC schemes and FHE.
This work offers system-level insights for researchers and practitioners seeking to understand or accelerate PPML workloads.