The design that dominated image work for about a decade. Instead of looking at a whole picture at once, it slides a small window across the image looking for a particular feature, then repeats with other windows looking for other features. Because the same window is reused everywhere, a CNN can spot a cat whether it appears top left or bottom right.
Technical view
Convolutional layers apply learned filters across the input with shared weights, giving translation invariance and far fewer parameters than a fully connected layer would need. Pooling layers progressively reduce spatial resolution while retaining the strongest signals. CNNs remain common in medical imaging and on constrained hardware, though transformer-based vision models now match or beat them at large scale, given enough data.
Related
Part of AI Terminology Index, section: Other Kinds of Neural Network.