PraxiMLP: A Threshold-based Framework for Efficient Three-Party MLP with Practical Security
Tianle Tao, Shizhao Peng, Haogang Zhu
TL;DR
PraxiMLP tackles the efficiency and precision challenges of privacy-preserving neural network training by introducing two novel additive-to-multiplicative and multiplicative-to-additive conversions that operate entirely within the arithmetic domain. This enables native floating-point support for nonlinearities like ReLU and Softmax, avoiding costly cross-domain conversions. The framework demonstrates dramatic precision gains and WAN-training speedups over mainstream PPML systems, validated on a three-party MLP across multiple datasets. Together, these contributions offer a practical pathway to deploy secure neural networks in real-world settings, while outlining directions for formal security definitions and broader architectural extensions.
Abstract
Efficiency and communication cost remain critical bottlenecks for practical Privacy-Preserving Machine Learning (PPML). Most existing frameworks rely on fixed-point arithmetic for strong security, which introduces significant precision loss and requires expensive cross-domain conversions (e.g., Arithmetic-to-Boolean) for non-linear operations. To address this, we propose PraxiMLP, a highly efficient three-party MLP framework grounded in practical security. The core of our work is a pair of novel additive-to-multiplicative conversion protocols that operate entirely within the arithmetic domain, thus avoiding expensive cross-domain conversions. By natively supporting loating-point numbers, PraxiMLP precisely handles non-linear functions, dramatically improving both efficiency and precision. Experimental results confirm that, compared to mainstream PPML frameworks, PraxiMLP delivers an average 8 orders of magnitude precision improvement on basic protocols and a 5x average model training speedup in a WAN environment.