Skip to content

SoK: Connecting the dots in privacy-preserving ML - systematization of MPC protocols and conversions between secret sharing schemes

Authors: M. Zbudila; A. Suresh; H. Yalame; O. Mirzamohammadi; A. Abidin; B. Preneel
Year: 2025

type: survey collection: sok-references status: auto

Links: ePrint 2025/1679

Summary

Auto-extracted summary (unreviewed)

This SoK gives a comparative overview of MPC frameworks for privacy-preserving ML, focusing on protocols that introduce novel approaches, and analyzes them by computational and communication complexity, throughput, security guarantees and suitability for small-party settings. It examines each PPML primitive from an MPC perspective, highlights the diversity of secret-sharing schemes and their interoperability problems, and proposes conversions between secret-sharing schemes to enable efficient hybrid solutions, accompanied by an open-source repository of primitive costs.

Abstract

Privacy-preserving machine learning (PPML) has become increasingly important due to the need to protect sensitive data during training and inference. Secure multiparty computation (MPC) and homomorphic encryption (HE) have emerged as foundational technologies, enabling secure computation over private data. In this work, we provide a systematic comparative overview of MPC frameworks for PPML, focusing on protocols that introduce novel approaches rather than incremental improvements. Frameworks are analyzed based on computational and communication complexity, throughput, security guarantees, and applicability in small-party settings. Each underlying primitive in PPML is examined from an MPC perspective, highlighting its role and trade-offs. We also emphasize the diversity of secret-sharing schemes and associated interoperability challenges, proposing scheme conversions to facilitate efficient hybrid solutions. This Systematization of Knowledge guides researchers in identifying open problems and practitioners in selecting effective MPC-based frameworks for real-world PPML deployment.
Source: pdf

Extracted details

Field Value Source Confidence
contribution Systematic comparative overview of MPC frameworks for PPML (training and inference) llm 0.70
contribution Analysis of each PPML primitive from an MPC perspective with its trade-offs llm 0.70
contribution Proposed conversions between secret-sharing schemes to enable hybrid protocols llm 0.70
contribution Guidance on open problems and framework selection; open-source repository of performance costs of PPML primitives llm 0.70
schemes secret sharing llm 0.70
schemes homomorphic encryption llm 0.70
tasks privacy-preserving ML training and inference llm 0.70
threat model multiple settings surveyed (semi-honest and malicious, small-party MPC) llm 0.70
interaction pure MPC llm 0.70
code url https://github.com/KULeuven-COSIC/sok-ppml-mpc-frameworks (URL found in the full text) regex 0.50
llm relevance medium llm 0.70

SoK reference [27]

M. Zbudila, A. Suresh, H. Yalame, O. Mirzamohammadi, A. Abidin, and B. Preneel, “SoK: Connecting the dots in privacy-preserving ML - systematization of MPC protocols and conversions between secret sharing schemes,” Cryptology ePrint Archive, Paper 2025/1679, 2025. [Online]. Available: https://eprint.iacr.org/2025/1679

Listed on the reference page of the site.

Keyword mentions

Counts of vocabulary terms in the full text (first page of occurrence in brackets).

  • scheme: MPC 184 [p.1], Secret sharing 103 [p.1], Function secret sharing 67 [p.2], Garbled circuits 13 [p.3], Zero-knowledge proofs 6 [p.5], Oblivious transfer 5 [p.5], TEE / confidential computing 5 [p.2], Paillier 4 [p.16], CKKS 3 [p.17], Differential privacy 3 [p.1], Leveled / somewhat HE 3 [p.15], BFV 2 [p.30], ElGamal 2 [p.15]
  • library: ABY / ABY3 23 [p.8], EzPC / CrypTFlow 10 [p.16], MP-SPDZ 2 [p.16], CrypTen 1 [p.16], TensorFlow 1 [p.16]
  • model: Transformer 29 [p.6], LLaMA 10 [p.15], GPT 5 [p.15], Decision tree 2 [p.16], ViT 1 [p.18], ResNet 1 [p.4], CNN 1 [p.17], RNN / LSTM 1 [p.17], Logistic regression 1 [p.34], Gradient boosting / XGBoost 1 [p.16]
  • hardware: GPU 6 [p.9]
  • technique: Two-party computation 73 [p.1], Three-party computation 57 [p.1], Malicious model 33 [p.4], Semi-honest model 32 [p.4], Convolution 29 [p.6], ReLU 28 [p.8], Lookup table 28 [p.9], Softmax 20 [p.8], Matrix multiplication 20 [p.5], Sigmoid 19 [p.8], GELU 18 [p.8], Rotations 13 [p.7], Non-interactive 12 [p.1], SIMD packing 7 [p.7], LayerNorm 6 [p.8], Attention 5 [p.8], Quantization 3 [p.15], Federated learning 3 [p.4], Fine-tuning 2 [p.18]
  • task: IoT / edge 4 [p.1], Medical / healthcare 3 [p.1], Finance 3 [p.1], Recommendation 2 [p.19]
  • topic: Privacy-preserving machine learning 102 [p.1], Homomorphic encryption 85 [p.1], Private / secure inference 13 [p.3], Large language models 9 [p.3], Open source 7 [p.2], Benchmark 5 [p.2]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/sok-references/027_sok-connecting-the-dots-in-privacy-preserving-ml-systematiza.pdf sok-references 43 712.8 KB 3505f3fb37e8

Database key: sok-connecting-the-dots-in-privacy-preserving-ml-systematization-of-mpc · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set sok-connecting-the-dots-in-privacy-preserving-ml-systematization-of-mpc COLUMN VALUE