Squeeze-and-Excitation Networks (SENets)

SENets are a building block for convolutional neural networks (CNNs)
the Squeeze-and-Excitation block (SE) is an architectural unit that can be plugged into a CNN to improve channel interdependencies between different feature channels
the SE block works by squeezing each channel into a single numeric value, which gives the block a global understanding of each channel
the SE block is designed to improve the representational power of a network