vertical-federated-learning

This paper systematically studies backdoor vulnerabilities in vertical federated learning (VFL), revealing a gap between research assumptions and real-world practice, and introduces BVBench, a backdoor-centric benchmark for practical evaluation.

本文研究垂直联邦学习中的选择性升级问题，提出一种基于期望增益的评分方法，在低成本的本地预测和高成本的嵌入融合之间进行路由，以优化通信-准确率权衡。