convolutional network

A convolutional neural network (CNN) is a neural-network architecture that employs local receptive fields and weight-sharing filters to process structured data, such as images, producing feature maps that are (approximately) translation-equivariant.

After early work like LeNet in handwriting recognition and the breakthrough AlexNet at the 2012 ImageNet challenge, CNNs became the dominant approach in vision. Variants now extend to audio, video and other grid-structured domains.

Compared to fully-connected networks, their local structure and parameter sharing improve sample efficiency, computational cost and generalisation on grid-like inputs.