1. Home
  2. Glossary
  3. Neural network
AI glossary · Foundations

Neural network

Neural network: A neural network is a computing system made of layers of simple connected units that pass numbers to each other, with connection strengths (weights) adjusted during training until the network produces useful output. It is the building block of modern AI.

Picture a grid of tiny calculators arranged in layers. Each one takes numbers from the layer before it, multiplies them by its own weights, adds them up, and passes the result forward. Input goes in one end (pixels, word pieces, sensor readings), and a prediction comes out the other (a label, a next word, a probability).

The magic is in training. The network starts with random weights and makes terrible guesses. Each guess is compared to the right answer, and the weights are nudged slightly to reduce the error. Repeat that millions of times and the network becomes good at the task, even though no human ever wrote the rules it follows.

The 'neural' name is a loose analogy to brain cells, not a claim that the software thinks like you do. What matters in practice is that a trained network is just a large collection of numbers, which is why the same architecture can be copied, fine-tuned, shrunk to run on a phone, or scaled up to billions of parameters.

Example at work

A retail chain's returns system uses a small neural network to score whether a return is likely fraudulent, based on purchase history, timing, and item type. The network was trained on past returns that investigators had labeled. Anything above a threshold goes to a human reviewer instead of being auto-approved.

Why it matters

You will hear 'neural network,' 'deep learning,' and 'model' used almost interchangeably. Recognizing that they all describe a trained set of weights, not a rulebook, explains why AI output can be brilliant on one input and nonsensical on a slightly different one: the network never learned that case.

Related terms