Perceptron

A perceptron is the simplest form of an artificial neural network. Introduced by Frank Rosenblatt in 1958, it consists of a single artificial neuron that takes multiple inputs, multiplies each by a weight, sums the results, and passes them through an activation function to produce a single output.

How a perceptron works

Think of a perceptron as a simple voting system:

1. Inputs: Each input represents a feature of the data (e.g., customer age, spending amount, account tenure).
2. Weights: Each input is multiplied by a weight that reflects its importance. Important features get larger weights.
3. Summation: All weighted inputs are added together, plus a bias term.
4. Activation: The sum is passed through a function that produces the output — typically a step function that outputs 1 if the sum exceeds a threshold and 0 otherwise.

If the output is 1, the perceptron classifies the input into one category. If 0, the other category.

Learning in a perceptron

A perceptron learns by adjusting its weights. When it makes a wrong prediction, the weights are updated to make the correct answer more likely next time. This update rule is one of the earliest examples of what we now call machine learning — the system improves through experience rather than explicit programming.

The XOR problem and its legacy

In 1969, Marvin Minsky and Seymour Papert published a book demonstrating that a single perceptron cannot solve certain problems — most famously the XOR (exclusive or) problem. This limitation contributed to the first "AI winter," a period of reduced funding and interest in neural networks.

The solution, discovered later, was to stack multiple layers of perceptrons together — creating a multi-layer perceptron (MLP). These deeper networks can solve problems that a single perceptron cannot. This insight eventually led to modern deep learning.

From perceptron to deep learning

Every neural network you encounter today — from image classifiers to large language models — is built from layers of units that are conceptual descendants of the perceptron. The basic principle remains the same: take inputs, apply weights, sum, and pass through an activation function. The difference is scale: modern networks have billions of these units arranged in sophisticated architectures.

Why this history matters

Understanding the perceptron helps you grasp why neural networks work the way they do. The concepts of weights, biases, activation functions, and learning rules introduced by the perceptron remain the fundamental vocabulary of deep learning sixty years later.