ReNet: A Recurrent Neural Network Based Alternative to Convolutional Networks

TL;DR

ReNet introduces a recurrent neural network architecture as an alternative to convolutional networks for object recognition, replacing convolution and pooling layers with recurrent sweeps across images, showing promising results on standard benchmarks.

Contribution

This paper presents ReNet, a novel deep neural network architecture that replaces convolutional layers with recurrent layers for image recognition tasks.

Findings

ReNet performs competitively on MNIST, CIFAR-10, and SVHN datasets.

ReNet demonstrates the viability of recurrent networks as an alternative to convolutional networks.

Further research is suggested to explore ReNet's potential and improvements.

Abstract

In this paper, we propose a deep neural network architecture for object recognition based on recurrent neural networks. The proposed network, called ReNet, replaces the ubiquitous convolution+pooling layer of the deep convolutional neural network with four recurrent neural networks that sweep horizontally and vertically in both directions across the image. We evaluate the proposed ReNet on three widely-used benchmark datasets; MNIST, CIFAR-10 and SVHN. The result suggests that ReNet is a viable alternative to the deep convolutional neural network, and that further investigation is needed.