Reproduce “EVA-S2PMLP: Secure and Scalable Two-Party MLP via Spatial Transformation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shizhao Peng, Shoumo Li, Tianle Tao
EVA-S2PMLP is a secure two-party computation framework designed for Multi-Layer Perceptron (MLP) training and inference with vertically-partitioned data, achieving near-plaintext accuracy and significantly reduced runtime, especially in high-latency WAN environments. The framework operates in the real-number domain, enabling exact implementations of non-linear activation functions and integrating lightweight verification mechanisms for computational integrity.
Key insights
- Secure computation performed in the real-number domain enables exact implementations of non-linear functions like ReLU and Softmax, thereby preserving model accuracy.
- Minimizing communication rounds, rather than just total volume, is critical for achieving high performance and scalability in high-latency WAN environments.
- Integrating lightweight verification mechanisms directly into secure protocols provides strong integrity guarantees without incurring substantial performance overhead.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.