Vertical Federated Learning Explainer
Vertical federated learning is used when organizations hold different columns, or features, about overlapping entities. One participant might have transaction history while another has device signals for some of the same customers.
The parties first identify matching entities through a privacy-aware alignment process. They then coordinate model calculations using encrypted values, secret sharing, split learning, or related protocols while keeping their raw feature tables separate.
This differs from horizontal federated learning, where participants usually hold the same feature schema for different rows. Vertical settings often require tighter coordination because a single prediction depends on features held by multiple parties.
The system must address identifier leakage, incorrect matches, collusion, dropout, labels held by one party, and legal limits on joint use. Keeping raw tables separate does not by itself establish privacy or authorization.
Vertical federated learning trains across organizations that hold different features about overlapping entities without pooling all raw records in one place.
Key terms include feature partition, overlapping entity, entity alignment, private set intersection, label holder, joint model, collusion, identifier and federation.
Each organization contributes distinct features for some of the same entities.
A coordinated protocol combines calculations without centralizing every raw record.
Matching records can reveal who appears in another participant's dataset.
Privacy, data-use authority, and matching errors must be handled explicitly.
Explore more "Explainers"
Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.