In 1958 the psychologist Frank Rosenblatt published “The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain” in Psychological Review (volume 65, issue 6, pages 386 to 408). It introduced one of the first artificial neural networks that could learn from examples.
The perceptron takes a set of inputs, weights them, and fires if the weighted total crosses a threshold, much like the McCulloch and Pitts neuron. Its key advance was a learning procedure: when the perceptron made a mistake, it automatically adjusted its weights to do better next time. This meant the machine could be trained to recognize patterns rather than having every rule programmed by hand.
Rosenblatt’s work, later built into physical hardware, generated enormous excitement and press coverage about thinking machines. It established the template, weighted inputs plus a learning rule, that underlies modern deep learning, even though its limitations would soon trigger a sharp backlash.