What is a neural network explained without the jargon

A Simple Idea Underneath the Complex-Sounding Name

Despite the intimidating name, a neural network is, at its core, a fairly straightforward kind of computing system built from many simple processing units connected together — the complexity people associate with neural networks comes from how many of these simple units are combined and trained, not from any single unit being individually sophisticated.

The Basic Building Block

Each individual processing unit, sometimes called a “node” or loosely a “neuron,” takes in some numerical inputs, combines them in a simple mathematical way, and produces an output. On its own, this single unit does something quite simple — the power of a neural network comes from connecting enormous numbers of these units together in an organized structure.

How the Units Are Organized Into Layers

These units are typically organized into layers, where the outputs of one layer become the inputs to the next. Information flows from an initial input layer, through one or more intermediate layers, to a final output layer, with each successive layer building a more refined or abstract representation of the original input based on what previous layers detected.

What ‘Weights’ Actually Are

Every connection between units in the network has an associated numerical value, called a “weight,” that determines how strongly that particular connection influences the next unit’s output. You can think of weights as adjustable dials controlling how much attention each connection pays to the signal passing through it.

What ‘Training’ Actually Does

Training a neural network means gradually adjusting all of these weights, based on many examples, so that the network’s final output gets closer and closer to a correct or desired result. This adjustment process happens automatically through a mathematical procedure that nudges each weight slightly in the direction that would have produced a better result for the examples seen so far, repeated over enormous numbers of examples until the network’s performance stabilizes.

Why It’s Called ‘Neural’ at All

The “neural” name reflects a loose, historical inspiration from simplified ideas about how biological brain cells connect and pass signals to each other — it’s a useful conceptual analogy for understanding the basic structure, but modern neural networks are mathematical systems that function quite differently from the vastly more complex biology of an actual brain.

Bottom Line

A neural network is a system of simple processing units, organized into connected layers, where adjustable numerical weights on each connection are gradually tuned during training so the network’s output gets progressively better at solving a given task — a conceptually simple building block whose real power comes from scale and training, not from any single part being complex.

Look Up AI Terms

Search plain-English definitions of AI and machine learning terms in our free AI Glossary.

Go deeper

• What Role Does AI Play in Supply Chain Network Design?