Learning Implicitly Recurrent CNNs Through Parameter Sharing
This work addresses the need for more parameter-efficient and flexible neural network architectures for researchers and practitioners in machine learning, though it is incremental as it builds on existing CNN and recurrent network concepts.
The paper tackled the problem of designing efficient convolutional neural networks (CNNs) by introducing a parameter sharing scheme that hybridizes CNNs with recurrent networks, resulting in substantial parameter savings while maintaining competitive accuracy on image classification tasks and improved training speed and extrapolation on algorithmic synthetic tasks.
We introduce a parameter sharing scheme, in which different layers of a convolutional neural network (CNN) are defined by a learned linear combination of parameter tensors from a global bank of templates. Restricting the number of templates yields a flexible hybridization of traditional CNNs and recurrent networks. Compared to traditional CNNs, we demonstrate substantial parameter savings on standard image classification tasks, while maintaining accuracy. Our simple parameter sharing scheme, though defined via soft weights, in practice often yields trained networks with near strict recurrent structure; with negligible side effects, they convert into networks with actual loops. Training these networks thus implicitly involves discovery of suitable recurrent architectures. Though considering only the design aspect of recurrent links, our trained networks achieve accuracy competitive with those built using state-of-the-art neural architecture search (NAS) procedures. Our hybridization of recurrent and convolutional networks may also represent a beneficial architectural bias. Specifically, on synthetic tasks which are algorithmic in nature, our hybrid networks both train faster and extrapolate better to test examples outside the span of the training set.