Skip Connections

In a encoder-decoder structure (Convolutional Neural Network ) it can be helpful to connect the symmetric layers to propagate high frequency source information to the decoder layers.

UNet Was one of the first to introduce it(?)