Perceptron
Early single-layer neural model that learns a linear decision boundary from examples; Rosenblatt (1958), later limited by Minsky and Papert's analysis.
A perceptron is a simple single-layer neural network unit: inputs with adjustable weights feed a threshold function for a binary yes/no decision. Frank Rosenblatt’s 1958 design showed that such a system could learn from errors by updating weights, a core idea behind modern machine learning.
The 1969 book Perceptrons by Marvin Minsky and Seymour Papert proved that one-layer perceptrons cannot represent some functions (notably XOR). Funding committees often read that as “neural nets fail,” though multi-layer networks could in principle escape the limits. Multi-layer training became practical with backpropagation in the 1980s; today’s deep learning stacks still rely on weight updates from prediction errors.