AI glossary · Fundamentals
Neural network
In one sentence
An artificial neural network is a mathematical model made of layers of connected units, called neurons, that turn a numeric input into an output. It learns by adjusting the weight of each connection based on examples.
Also known as: artificial neural network · ANN · neural net
What is a neural network?
The name suggests a brain, but the reality is more down to earth. A neural network is a chain of mathematical operations organised in layers. Each unit, or neuron, takes in numbers, combines them and passes the result on to the next layer.
Put it to work
AI agents that look after your customers
From €49/mo + VAT
Voice, WhatsApp and email with the phone line included: agent, numbers and CRM in one fee.
See plans and pricingWhat makes it useful is that nobody tells it exactly what calculation to do. The network starts with random values and keeps adjusting them against examples until its output looks like the right answer. When it has many layers, it’s called deep learning.
Related to this term
UTIO agents
- Available 24/7 in 26 languages
- Set it up yourself for €0: built from your website
- 14-day free trial, no card needed
From €49/mo + VAT
See plans and pricingHow it works
- Input layer. Takes in data already turned into numbers: an image, a recording or the tokens of a text.
- Hidden layers. Each neuron multiplies its inputs by a set of weights, adds them up and applies a function that decides how strongly it “fires”. The result moves on to the next layer.
- Output layer. Returns the answer: the probability of each category, a number or the next token.
The weights are the network’s parameters, and they hold everything it has learned. During training, the network compares its output with the right answer, measures the error and passes it backwards, layer by layer, to tweak each weight slightly. That’s called backpropagation. Repeated millions of times, it brings the error down.
Different architectures suit different data. For text, the dominant one today is the transformer, which lets the network relate every word to every other word in a sentence.
Example
A language school in Edinburgh wants to spot emails from students who are about to drop out, so it can reach them first. A simple neural network takes each email, converted to numbers, and returns a probability that it’s a cancellation.
At first it’s guessing. After training on a couple of years of emails that staff had already labelled, it learns that certain phrasing is a strong signal, even when the student never says “cancel”. The school didn’t write a single rule: they live in the network’s weights.
What it means for your business
Knowing what a neural network is explains two traits of any modern AI. First, it learns from examples, so its quality depends on the data it was trained on. Second, its “reasoning” is spread across millions of weights, so it can’t justify an answer the way a person would.
That’s why, in UTIO’s agents, the network never works without guardrails you set: you decide what the voice agent can do, which information it answers from and when it hands over to a person. And it starts in draft mode until you’ve seen that its replies are reliable.
Frequently asked questions
Does a neural network work like the human brain?
Only very loosely. Each artificial 'neuron' is a simple maths operation: it multiplies its inputs by weights, adds them up and applies a function. There's no biology and no thinking, just a great deal of calculation in parallel.
What types of neural network are there?
The best known are dense networks (every neuron linked to every neuron in the next layer), convolutional networks, widely used for images, recurrent networks, designed for sequences, and transformers, the architecture behind today's language models.
What is a hidden layer?
Any layer between the input and the output. It's called hidden because you don't directly see what it computes: these are intermediate representations the network learns for itself during training.
Related terms
- Deep learning
Deep learning is a type of machine learning that uses neural networks with many layers. Each layer picks out more abstract features of the data, which lets it handle text, speech and images without hand-written rules.
- Model parameters
A model's parameters are its internal numeric values, mainly the weights of a neural network, which are adjusted during training and store what it has learned. Their count is used as a rough measure of model size.
- Transformer
A transformer is a neural network architecture that processes every word in a text at once and uses an attention mechanism to work out which parts relate to each other. It's the foundation of almost every language model in use today.
- Model training
Model training is the process in which an AI model learns from a dataset: it's shown examples, its errors are measured and its parameters are adjusted, over and over, until its outputs are good.
- Machine learning
Machine learning is the branch of AI in which a system learns to do a task from examples, rather than following hand-written rules, and gets better as it sees more data.
Updated on