Deep Residual Learning for Image Recognition

Authors: He, K., Zhang, X., Ren, S., & Sun, J. Year: 2015 ArXiv: https://arxiv.org/abs/1512.03385

Summary

ResNet introduces residual connections (skip connections) that allow training of very deep networks (152+ layers). It achieves unprecedented accuracy on ImageNet and becomes the foundation for modern computer vision architectures.

Key Concepts

• Skip connections for gradient flow
• Residual learning framework
• Batch normalization for stability
• Very deep networks (50, 101, 152 layers)
• Identity mappings between layers

Impact

ResNet solved the degradation problem in deep networks and enabled training of networks with 100+ layers. It's the most influential architecture in computer vision since AlexNet and remains widely used.

Related Papers

• AlexNet: ImageNet Classification
• VGG: Very Deep Convolutional Networks
• Batch Normalization

---

Citation:

@inproceedings{he2015deep,
  title={Deep Residual Learning for Image Recognition},
  author={He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian},
  booktitle={IEEE Conference on Computer Vision and Pattern Recognition},
  pages={770--778},
  year={2015}
}