Do deep convolutional nets really need to be deep and convolutional?

Do deep convolutional nets really need to be deep and convolutional?

Urban, Gregor and Geras, Krzysztof J. and Kahou, Samira Ebrahimi and Aslan, Ozlem and Wang, Shengjie and Mohamed, Abdelrahman and Philipose, Matthai and Richardson, Matt and Caruana, Rich

5th International Conference on Learning Representations, ICLR 2017 – Conference Track Proceedings 2017

Abstract : Yes, they do. This paper provides the first empirical demonstration that deep convolutional models really need to be both deep and convolutional, even when trained with methods such as distillation that allow small or shallow models of high accuracy to be trained. Although previous research showed that shallow feed-forward nets sometimes can learn the complex functions previously learned by deep nets while using the same number of parameters as the deep models they mimic, in this paper we demonstrate that the same methods cannot be used to train accurate models on CIFAR-10 unless the student models contain multiple layers of convolution. Although the student models do not have to be as deep as the teacher model they mimic, the students need multiple convolutional layers to learn functions of comparable accuracy as the deep convolutional teacher.