A convolutional neural network (CNN) is a type of neural network built for grid-like data such as images. Instead of connecting every pixel to every neuron, it slides small filters across the image to detect local features like edges and textures, then combines them into higher-level shapes. Because the same filter is reused everywhere, a CNN can recognize a feature no matter where it appears in the picture.
The core idea traces to Kunihiko Fukushima’s 1980 “Neocognitron,” published in Biological Cybernetics, volume 36, a self-organizing network designed to recognize patterns unaffected by shifts in position. The modern trainable form was established by Yann LeCun and colleagues in the 1998 paper “Gradient-based learning applied to document recognition” in Proceedings of the IEEE, which introduced the LeNet architecture and applied it to reading handwritten digits.
CNNs are the foundation of computer vision. They powered the breakthrough image-recognition results of the 2010s and remain central to applications from medical imaging to self-driving cars.
Why business readers should care: any product that “sees,” such as document scanning, defect detection on a production line, facial recognition, or photo tagging, almost certainly relies on the convolutional approach pioneered by these papers.