VHFed: A Two-Tier Vertical and Horizontal Federated Learning Framework for Enhanced Model Performance

Federated Learning (FL) enables collaborative model training while preserving data privacy, making it a critical solution for sensitive data scenarios. However, Vertical Federated Learning (VFL) faces challenges such as inefficient data utilization, high communication costs, and limited personalization in heterogeneous data environments. This paper proposes VHFed, a two-tier hybrid federated learning framework combining horizontal and vertical FL to address these challenges. VHFed incorporates a global local neural network (GLNN) to support personalized learning while preserving participant privacy. Theoretical analysis and experiments on multiple datasets demonstrate that VHFed significantly improves accuracy, personalization, and privacy protection, showcasing its advantages in handling heterogeneous data and multi-party collaborations.