A Perceptron is the simplest form of a neural network unit that applies a weighted sum of inputs and an activation function. It was one of the earliest neural models, laying the foundation for multi-layer networks. Single-layer perceptrons can only learn linearly separable tasks.