Federated learning is a decentralized approach to training machine learning (ML) models. Each node across a distributed network trains a global model using its local data, with a central server aggregating node updates to improve the global model.