Federated learning for the Internet of Things (IoT) is a rapidly growing research area that focuses on enabling decentralized, privacy-preserving model training across distributed IoT devices, which often generate sensitive or heterogeneous data. This paradigm allows collaborative learning without transmitting raw data to central servers, addressing privacy, bandwidth, and regulatory constraints. Research explores optimization techniques for resource-constrained IoT devices, handling non-i.i.d. and unbalanced data, communication-efficient aggregation, model compression, and personalization strategies to improve local and global model performance. Integration with deep learning architectures, such as CNNs, RNNs, and transformers, supports applications including anomaly detection, predictive maintenance, smart home automation, healthcare monitoring, and intelligent transportation systems. Recent studies also investigate secure aggregation, differential privacy, blockchain-based incentives, and robustness against adversarial attacks, establishing federated learning as a key enabler for scalable, secure, and intelligent IoT ecosystems.