Reproduce “Ents: An Efficient Three-party Training Framework for Decision Trees by Communication Optimization” 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
Guopeng Lin, Weili Han, Wenqiang Ruan, Ruisheng Zhou, Lushan Song, Bingshuai Li, Yunfeng Shao
The Ents framework introduces an efficient three-party training method for decision trees, addressing communication bottlenecks in secure multi-party computation. It achieves 5.5x to 9.3x lower communication and 3.9x to 5.3x fewer communication rounds compared to prior state-of-the-art methods, enabling training on over 245,000 samples in under three hours in a WAN setting while maintaining accuracy.
Key insights
- By leveraging the stability of radix sort and updating pre-generated permutations, Ents combines the benefits of one-time permutation generation with linear communication growth, avoiding prior inefficiencies.
- Separating computations based on their required bit-length and performing share conversion only when necessary significantly reduces the overall communication footprint.
- The `ConvertShare` protocol efficiently handles positive secret values from a smaller ring to a larger ring with minimal communication and rounds by focusing on overflow correction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.