Perceptron

Neural Networks started in the form of perceptron introduced by Rosenblatt which included a single linear mapping and an activation function, originally the step function.

Rosenblatt's Perceptron was originally conceived as a three-component system for complex pattern recognition:

• S-System (Sensory System): The input (e.g. a set of points in photocells)
• A-System (Association System): A set of A-units (the actual Perceptron units) that perform the weighted sum and thresholding, acting as the switching and memory function.
• R-System (Response System): The output layer, which measures the aggregated mean value from the A-system and triggers a response (like a signal light or printer).

A single A-unit is the single-layer perceptron unit often referenced today, performing the linear mapping and step function, but the complete Perceptron system described in the 1957 report was an entire architecture designed to learn and recognize patterns.