Deep representation-based transfer learning for deep neural networks

Highlights：

• A deep representation-based transfer learning method is proposed for knowledge transfer between deep neural networks.

• The method transfers domain-invariant representation by learning the similarity structure between source representations.

• The method reduces the influence of unfavorable source representations through boosting-based representation transfer.

• The computational process of the method is vectorized with comparable training efficiency to traditional fine-tuning.

• The image classification and time series prediction experiments demonstrate the universal transfer capability of the method.