Das Neocognitron is a hierarchical, multilayered künstliches neuronales Netzwerk introduced by Kunihiko Fukushima in the 1980s. It is primarily designed for visual pattern recognition, enabling machines to identify and classify images effectively.
Der Kern architecture of the Neocognitron is inspired by the human visual system. It consists of multiple layers of neurons that process visual input in a hierarchical manner. Each layer performs specific functions: initial layers detect simple features, while subsequent layers capture more complex patterns by combining features from the previous layers. This hierarchical processing is akin to how humans perceive and interpret visual stimuli.
The Neocognitron employs convolutional operations, similar to those found in modern Konvolutionale Neuronale Netze (CNNs), allowing it to efficiently handle spatial hierarchies in images. Additionally, the network utilizes a unique approach to learning known as kompetitives Lernen, where neurons compete to respond to particular features, effectively enhancing the network’s ability to generalize from training examples.
Despite being an early model in the field of neural networks, the Neocognitron laid the groundwork for subsequent advancements in deep learning and computer vision. Its principles continue to influence the design of contemporary neural networks and have contributed to significant developments in areas such as image recognition, Objekterkennung, and more.