FedCVT: Semi-supervised Vertical Federated Learning with Cross-view Training

Federated learning allows multiple parties to build machine models collaboratively without exposing data. In particular, vertical federated (VFL) enables participating a joint model based upon distributed features of aligned samples. However, VFL requires all share sufficient amount reality, the set samples may be small, leaving majority non-aligned data unused. this article, we propose Cross-view Training (FedCVT), semi-supervised approach that improves performance with limited More specifically, FedCVT estimates representations for missing features, predicts pseudo-labels unlabeled expand training set, and trains three classifiers jointly different views expanded improve model's performance. does not require their original parameters, thus preserving privacy. We conduct experiments on NUS-WIDE, Vehicle, CIFAR10 datasets. The experimental results demonstrate significantly outperforms vanilla only utilizes Finally, perform ablation studies investigate contribution each component FedCVT.