Randomness in Deconvolutional Networks for Visual Representation

Kun He, Jingbo Wang, Haochuan Li, Yao Shu, Liwei Wang, John E. Hopcroft

TL;DR: We investigate the deep representation of untrained, random weight CNN-DCN architectures, and show their image reconstruction quality and possible applications.

Abstract: To understand the inner work of deep neural networks and provide possible theoretical explanations, we study the deep representations through the untrained, random weight CNN-DCN architecture. As a convolutional AutoEncoder, CNN indicates the portion of a convolutional neural network from the input to an intermediate convolutional layer, and DCN indicates the corresponding deconvolutional portion. As compared with DCN training for pre-trained CNN, training the DCN for random-weight CNN converges more quickly and yields higher quality image reconstruction. Then, what happens for the overall random CNN-DCN? We gain intriguing results that the image can be reconstructed with good quality. To gain more insight on the intermediate random representation, we investigate the impact of network width versus depth, number of random channels, and size of random kernels on the reconstruction quality, and provide theoretical justifications on empirical observations. We further provide a fast style transfer application using the random weight CNN-DCN architecture to show the potential of our observation.

Keywords: Deep representation, random representation, untrained deconvolutional network, image reconstruction