Developing a Face Recognition System Using Vision Transformer and Federated Learning

Traditional centralized approaches to training AI systems for face recognition raise significant privacy concerns by consolidating massive datasets in a single location. This method exposes personal data to potential breaches and misuse. This paper proposes a Federated Learning framework utilizing a Vision Transformer model for face recognition with minimal local data. This framework ensures data security by facilitating collaborative training across multiple devices, thereby avoiding the risks associated with centralized data storage. Experimental results validate the effectiveness and reliability of our approach: the Vision Transformer model achieves stable performance when trained on limited local data using Federated Learning. This method supports efficient and secure data distribution, offering a promising solution to the privacy-performance trade-off in AI system development for face recognition while maintaining high performance and security standards.