Convolutional Neural Network (CNN)

A class of artificial neural network commonly used to analyze visual imagery.

How it works

CNNs apply a set of learned filters (kernels) across an input image using a sliding-window convolution operation. Each filter detects a specific low-level feature (an edge, a colour gradient) and produces a feature map. Stacking many convolutional layers allows the network to compose simple features into complex ones: edges become shapes, shapes become objects. Pooling layers then reduce the spatial dimensions, making the representation progressively more abstract and position-invariant.

Why it matters

CNNs were the breakthrough architecture that made modern computer vision possible. ImageNet victories starting in 2012 demonstrated that deep CNNs dramatically outperformed hand-engineered feature extraction. They power the object detection systems in smartphones, autonomous vehicles, and medical imaging tools. Their parameter efficiency through weight sharing — one filter applied everywhere — also made them practical to train at the time when GPU memory was far more limited than today.