Deep learning
Convolution
A convolution slides a shared filter across local regions of an input.
How it works
Weight sharing lets one detector recognize the same pattern at different positions. Local connectivity also uses fewer parameters than a dense layer over the full image.
A concrete example
Illustrative: a 3×3 filter over a 32×32 image with valid padding produces a 30×30 map.
Do not confuse it with Pooling
convolution learns filters; pooling summarizes nearby values without learned kernels.