A Secure Framework in Vertical and Horizontal Federated Learning Utilizing Homomorphic Encryption

To ensure data security and model training among different institutions, this paper combines vertical federated learning and horizontal federated learning. It utilizes homomorphic encryption to encrypt feature data and model weights, designing a secure framework for vertical and horizontal federated learning. This framework is suitable for binary classification applications. It enables various institutions within the same field, possessing the same feature data, and institutions from different field with identical client samples, to securely conduct federated learning training without exposing their privacy data.