The proposed framework achieved the state-of-the-art diagnosis accuracy of 97.79% and 94.84% on two bearing datasets with just five labeled samples. The result shows that using deep networks to assist ...