Connecting the Dots in
Privacy-Preserving ML

About the SoK

This systematization organizes MPC-based PPML frameworks across key dimensions: algebraic structures, threat models, execution phases, deployment modes, and network configurations.

We decompose PPML systems into core cryptographic primitives and provide theoretical cost analyses for all major neural network operations — from dot products and truncation to ReLU, GeLU, and normalization.

A central contribution is the MPC Puzzle — a unification of 2-, 3-, and 4-party secret-sharing schemes with a complete mapping of conversion protocols and their overhead.

BibTeX Citation

@misc{zbudila2025sok,
  title  = {{SoK}: Connecting the Dots in Privacy-Preserving {ML} ---
             Systematization of {MPC} Protocols and Conversions
             Between Secret Sharing Schemes},
  author = {Zbudila, Martin and Suresh, Ajith and Yalame, Mohammad Hossein
             and Mirzamohammadi, Saeid and Abidin, Aysajan and Preneel, Bart},
  year   = {2025},
  note   = {Cryptology ePrint Archive, Report 2025/1679}
}