Stacking combines multiple heterogeneous models (e.g., SVM + KNN + Tree) using a meta-model that learns how to best combine their predictions.
Advanced ensemble technique.
Used to squeeze the last few percent of performance in competitions.