Federated learning for smart city applications is an emerging research area that focuses on enabling decentralized, privacy-preserving learning across multiple urban systems and IoT devices, without sharing raw data. This paradigm supports collaborative intelligence for various smart city services, including traffic management, energy optimization, public safety, environmental monitoring, and healthcare, while addressing data privacy, regulatory compliance, and heterogeneous data challenges. Research explores integrating federated learning with deep learning architectures such as CNNs, RNNs, transformers, and graph neural networks (GNNs) to handle spatial–temporal data, sensor networks, and large-scale urban datasets. Studies also investigate communication-efficient model aggregation, personalization for local environments, secure and privacy-preserving techniques (e.g., differential privacy, homomorphic encryption), and robustness against adversarial attacks. Recent developments highlight applications in real-time traffic congestion prediction, energy demand forecasting, anomaly detection in utilities, and predictive public health monitoring, establishing federated learning as a key enabler for intelligent, scalable, and privacy-aware smart city systems.