Dynamic ensemble of regression neural networks based on predictive uncertainty

Yoonhyung Lee, Seokho Kang

Abstract: Highlights•We address decentralized environment where data are distributed across local nodes.•Dynamic ensemble of locally trained neural networks is constructed.•Access to data from local nodes is not required.•Prediction for query instance is made by combining outputs of neural networks.•Larger weights are assigned to neural networks with higher predictive uncertainty.