Short Definition
A convolution is an operation that applies a small learned filter across local regions of data.
Intuition
The same detector slides over an input and looks for local patterns wherever they appear.
Technical Definition
In neural networks, convolution applies shared kernels over spatial, temporal, or sequence dimensions to produce feature maps with local receptive fields.
Example
An image model may use convolutions to detect edges, textures, and shapes.
Common Misunderstandings
Convolutions are useful beyond images, including audio and sequences.
A convolutional layer is not automatically translation invariant; architecture and pooling matter.