ResNet: Enabling Deep Convolutional Neural Networks through Residual Learning

TL;DR

This paper discusses Residual Networks (ResNet), which use skip connections to enable the training of very deep CNNs, improving accuracy and training stability on image classification tasks.

Contribution

It demonstrates how residual learning allows training of hundreds of layers in CNNs, overcoming vanishing gradient issues and improving performance.

Findings

ResNet-18 achieves 89.9% accuracy on CIFAR-10.

ResNet trains faster and more stably than traditional deep CNNs.

Residual learning enables training of much deeper networks.

Abstract

Convolutional Neural Networks (CNNs) has revolutionized computer vision, but training very deep networks has been challenging due to the vanishing gradient problem. This paper explores Residual Networks (ResNet), introduced by He et al. (2015), which overcomes this limitation by using skip connections. ResNet enables the training of networks with hundreds of layers by allowing gradients to flow directly through shortcut connections that bypass intermediate layers. In our implementation on the CIFAR-10 dataset, ResNet-18 achieves 89.9% accuracy compared to 84.1% for a traditional deep CNN of similar depth, while also converging faster and training more stably.