Horizontal Federated Learning

Horizontal Federated Learning, or sample-based federated learning, is a type of Federated Learning where multiple parties collaborate to train a shared machine learning model without directly exchanging their raw data. In Horizontal Federated Learning, different parties possess different samples (or data points) but share the same feature set.

Example

Let's consider a simple example involving multiple hospitals that want to build a machine-learning model to predict patient readmission rates. Each hospital has collected data on various patient attributes like age, blood pressure, medical history, etc., which constitute the features. While each hospital has its own set of patients (samples), the attributes they collect (features) are the same across all hospitals.

How it Works

1. Initialization. A global model is initialized, often on a centralized server.
2. Local Training. Each party trains the global model using its local dataset, generating a local model update (e.g., gradient updates).
3. Model Aggregation. All local updates are sent to the centralized server, aggregating them to produce an updated global model.
4. Iteration. Steps 2 and 3 are iteratively repeated until the model converges or meets other stopping criteria.

Key Advantages

1. Data Privacy. Since raw data doesn't leave the local premises, the risk of data leakage is significantly reduced.
2. Efficiency. Horizontal Federated Learning enables efficient utilization of decentralized data.