Architectures · intermediate · concept 60 of 176
Residual Connections & ResNet
A shortcut that adds a layer's input to its output, letting gradients flow through very deep networks instead of vanishing. ResNet used it to jump from 20-layer to 150-layer networks in 2015, and every transformer block since is built around the same skip connection.
Key terms
Skip connectionIdentity mappingVanishing gradientsDeep networksResNet
Learn these first
Where you meet it in the real world
Inside every modern architecture: CNNs, transformers, diffusion U-Nets, all of them
Videos
▶ C4W2L03 Resnets ↗
DeepLearningAI · YouTube
▶ [Classic] Deep Residual Learning for Image Recognition (Paper Explained) ↗
Yannic Kilcher · YouTube
Guides and articles
8.6. Residual Networks (ResNet) and ResNeXt — Dive into Deep Learning 1.0.3 documentation ↗
Dive into Deep Learning
[1512.03385] Deep Residual Learning for Image Recognition ↗
arXiv (original paper)