Convolutional neural network
Also: CNN, ConvNet
A network built from convolution layers, where each small filter is reused at every position of the input, usually interleaved with pooling and followed by fully connected layers. Weight sharing and locality make CNNs efficient and translation-equivariant, which suits images.
output size = ⌊(W − K + 2P) / S⌋ + 1
Example
A 3×3 filter on a 32×32 input with no padding and stride 1 gives a 30×30 feature map; on an RGB image that filter has only 3·3·3 + 1 = 28 parameters, whatever the image size.