The problem with ranking ensembles based on training or validation performance.

Ulf Johansson, Tuve Lfstrm, Henrik Bostrm

Browse the full IJCNN paper archive.