In federated learning, a global model is a shared machine learning model that is collaboratively trained across multiple decentralized devices or servers without centralizing their raw data. Instead of sending data to a central server, each participant (e.g., a smartphone, IoT device, or local server) trains a local model on its own data. These local updates are then aggregated to create the global model, which serves as the unified representation of knowledge learned from all participants. For example, in a smartphone keyboard app, individual devices might train on users’ typing patterns locally, and only the model updates (not the actual keystrokes) are combined into a global model that improves autocorrect for all users.
The global model is built through an iterative process. First, a central server initializes a base model (e.g., a neural network with random weights) and distributes it to participants. Each device trains this model using its local data, typically via methods like stochastic gradient descent, and sends the updated model parameters back to the server. The server then aggregates these updates—often by averaging parameters (a technique called Federated Averaging, or FedAvg)—to produce a new global model. This cycle repeats, refining the global model over time. For instance, in a healthcare scenario, hospitals might train on patient records locally, and the server combines their updates to create a global diagnostic model without exposing sensitive data.
The global model’s key advantage is enabling privacy-preserving collaboration, as raw data never leaves local devices. It also scales well to large, distributed systems. However, challenges include handling non-identically distributed (non-IID) data across devices—for example, if some smartphones have mostly English text data while others have Spanish—which can bias the global model. Communication efficiency is another concern, as frequent updates between devices and the server can strain resources. Despite these challenges, the global model remains foundational in federated learning, offering practical solutions for industries like healthcare, finance, and IoT, where data privacy and regulatory compliance are critical.