Deep Collaborative Weight-based Classification.

One of the biggest problems in deep learning is its difficulty to retainconsistent robustness when transferring the model trained on one dataset toanother dataset. To conquer the problem, deep transfer learning was implementedto execute various vision tasks by using a pre-trained deep model in a diversedataset. However, the robustness was often far from state-of-the-art. Wepropose a collaborative weight-based classification method for deep transferlearning (DeepCWC). The method performs the L2-norm based collaborativerepresentation on the original images, as well as the deep features extractedby pre-trained deep models. Two distance vectors will be obtained based on thetwo representation coefficients, and then fused together via the collaborativeweight. The two feature sets show a complementary character, and the originalimages provide information compensating the missed part in the transferred deepmodel. A series of experiments conducted on both small and large visiondatasets demonstrated the robustness of the proposed DeepCWC in both facerecognition and object recognition tasks.