Convolutional neural network (CNN)
Neural network architecture for grid-like data such as images, using shared local filters and pooling to learn spatial features efficiently.
A convolutional neural network (CNN) is a neural network designed for data with spatial structure, especially images. Instead of connecting every input pixel to every unit in the first layer, it applies small convolutional filters repeatedly across the image, sharing weights and learning local features such as edges and textures. Training adjusts those filters end to end through backpropagation, often with pooling layers that downsample activations and improve robustness to small shifts.
Yann LeCun and colleagues at AT&T Bell Laboratories demonstrated the approach in 1989 on handwritten US Postal Service zip codes, an early proof that convolution plus gradient-based learning could scale to real recognition tasks. The same architectural family later powered AlexNet and the modern deep learning boom in computer vision. See the LeCun zip code article and timeline entry.