# Convolutional Neural Network **Domain:** Machine Learning / Computer Vision **Doc Type:** Canonical Architecture Node **Maturity:** Foundational ## Definition A **convolutional neural network**, or **CNN**, learns spatially organized features through shared local transformations, layered representation and aggregation across neighboring positions. ## Historical Lineage [[wiki/Kunihiko Fukushima|Kunihiko Fukushima's]] [[wiki/Neocognitron|neocognitron]] established a direct architectural ancestor through alternating feature-sensitive and pooling cells that produced position-tolerant recognition. ## Bell Labs Development [[articles/Bell Labs and the Distributed Architecture of American Power|Bell Labs and the Distributed Architecture of American Power]] documents [[wiki/Yann LeCun|Yann LeCun's]] development of backpropagation-trained convolutional networks at [[wiki/Bell Labs|Bell Labs]] during the late 1980s and 1990s, where the LeNet architecture achieved practical handwriting recognition for check processing. LeCun's Bell Labs work connected Fukushima's architectural insight to gradient-based learning, producing the lineage that leads directly to modern deep learning and LeCun's subsequent role as chief AI scientist at [[wiki/Meta|Meta]]. ## Technical Context Convolutional structure reuses detectors across a field, reducing parameter count while preserving local relationships. Pooling or related aggregation operations provide controlled tolerance to translation. ## See Also [[wiki/Computer Vision|Computer Vision]], [[wiki/Deep Learning|Deep Learning]], [[wiki/Representation Learning|Representation Learning]], [[wiki/Yann LeCun|Yann LeCun]], [[wiki/Bell Labs|Bell Labs]]