ML//CNN//ResNet

ResNet is the CNN that introduced skip connections, adding the input of each block directly to its output, \(x+f(x)\).

ResNet is the CNN that introduced skip connections, adding the input of each block directly to its output, x+f(x)x+f(x)x+f(x).

The skip connections allowed training networks of more than 100 layers (ResNet-152), and ResNet won ImageNet 2015 with superhuman accuracy. Transformers borrowed the same idea directly as the residual connection.