A part of Coaley Peak®Neural Network
A computing system inspired by the human brain, made up of connected nodes that process information in layers.
A neural network is a type of computing system modelled, at a high level, on the way neurons connect in biological brains. It consists of layers of simple processing units called nodes or neurons, each of which receives numerical inputs, applies a mathematical function to them, and passes the result onward. On its own, a single node does almost nothing useful. Assembled into layers and trained on data, a network of millions of these nodes can learn to recognise speech, translate languages, and generate images, performing tasks that once seemed to require human intelligence.
Neural networks are the foundation of deep learning and, by extension, most modern artificial intelligence. Understanding how they learn gives you a clearer picture of why today's AI behaves the way it does, including both its impressive capabilities and its characteristic failure modes.
How it works#
A neural network is organised into three main sections. The input layer receives raw data: pixel values from an image, numerical features from a spreadsheet, or tokenised words from a piece of text. One or more hidden layers process and transform that data, with each node in a layer connected to nodes in the layers before and after it. The output layer produces the final result: a classification, a prediction, a sequence of words, or something else depending on the task.
The connections between nodes have associated weights, which are numbers that control how much influence one node has on the next. During training, the network is shown many examples with known correct answers. It makes a prediction, measures how wrong it was, and then adjusts the weights slightly to do better next time. This adjustment process, repeated across millions of examples, is what training means. After training, the weights are fixed and the network is deployed to make predictions on new data.
The transformer architecture, introduced in 2017, is a particularly influential type of neural network that underlies most of today's large language models and many image generation systems.
Examples#
Facial recognition. Photo applications that automatically group pictures of the same person use neural networks. The network is trained on millions of labelled face images, learning to extract distinctive features such as eye spacing, nose shape, and jaw contour. When it sees a new photo, it matches those features against what it has learned.
Credit scoring. Some UK lenders use neural networks as part of their credit assessment process. The network is trained on historical data from past applicants, including whether they repaid their loans, and learns which combinations of factors are predictive of repayment. This can surface patterns in the data that a traditional scoring model built from hand-crafted rules might miss.
Why it matters#
Neural networks are the mechanism by which machine learning achieves its most impressive results. When a news headline describes a breakthrough in AI, a new neural network architecture is usually involved. For business users, the practical implication is that AI tools built on neural networks can generalise from examples in ways that older software cannot, making them far more adaptable to novel inputs. The trade-off is that they are less predictable and can fail in unexpected ways, which is why human oversight remains important.
Learn how neural networks are applied in language models by reading our tutorial How LLMs Work, or explore the broader picture in AI vs ML vs Deep Learning.
Neural Network: frequently asked questions
- Is a neural network the same as a human brain?
- No. Neural networks are loosely inspired by the structure of biological brains, but the resemblance is superficial. Biological neurons are extraordinarily complex chemical and electrical systems; artificial neurons are simple mathematical functions. The brain metaphor is useful for building intuition, but neural networks operate on very different principles.
- What is the difference between a neural network and deep learning?
- Deep learning is the practice of training neural networks that have many layers, often dozens or hundreds. A neural network with only one or two hidden layers is not typically called deep learning. In common usage today, 'neural network' usually implies a deep one, but strictly speaking a neural network can have any number of layers.
- Do neural networks need a lot of data to work well?
- Generally, yes. Neural networks with many layers have a large number of parameters to tune, and they need substantial amounts of training data to set those parameters reliably. With too little data, the network tends to memorise the training examples rather than learning generalisable patterns. Techniques like transfer learning can reduce this requirement by starting from a network already trained on a large dataset.
Go deeper with related tutorials
AI vs Machine Learning vs Deep Learning: What's the Difference?
Understand the difference between artificial intelligence, machine learning, and deep learning. Clear definitions with real-world examples.
Read tutorialHow Large Language Models Work (Simple Explanation)
A clear explanation of how large language models like the AI assistants people use every day actually work. No technical background needed.
Read tutorialUseful? Now picture your whole team knowing it.
Everything here is free, and the first course is too: eight short online lessons your team can start this afternoon. When you want the skills to stick, we come to you for a day on-site, built around the work your people actually do.