The Perceptron is widely regarded as the first implementation of a neural network, demonstrated by Frank Rosenblatt in 1957.
Rosenblatt developed the system at the Cornell Aeronautical Laboratory. The original Mark I Perceptron was an electronic machine inspired by the idea that networks of simple processing units could learn to classify patterns. It adjusted numerical weights when its predictions were wrong.
The perceptron could learn linearly separable classifications, such as separating inputs belonging to two classes with a straight decision boundary. Its limitations became especially important when researchers showed that a single-layer perceptron could not represent some simple functions, including exclusive OR.
Later multilayer neural networks and training methods overcame many of those limitations. The modern use of “neural network” covers much more complex architectures, but the perceptron remains a foundational model because it introduced an influential learning rule based on weighted inputs and corrective updates.