Neural network
A neural network is a computational model whose structure is loosely inspired by the human brain. It consists of many interconnected artificial neurons organised in layers that process information. Through training, the network learns to map inputs to fitting outputs and thus forms the technical basis for machine learning and deep learning.
Also known as: artificial neural network, neural network, ANN
How is a neural network structured?
A neural network consists of three kinds of layer: an input layer, one or more hidden layers and an output layer. Each layer contains artificial neurons connected to the neurons of the neighbouring layers. Every one of those connections has a weight stating how strongly a signal is passed on.
A single neuron sums the weighted inputs, adds a threshold value and applies a so-called activation function to it. That decides whether and how strongly the neuron passes a signal to the next layer. Only the interplay of many such neurons lets the network represent complex, non-linear relationships too.
How does a neural network learn?
A neural network learns by adjusting the weights of the connections so the outputs match the desired results as closely as possible. At the start the weights are set at random, so the network does not yet give meaningful predictions.
During training the network compares its predictions with the correct answers and calculates an error value from that. Through backpropagation the error is passed backwards through the network, and an optimisation method adjusts the weights step by step. After many repetitions across large amounts of data, the network gives increasingly accurate results.
What types of neural network are there?
Numerous variants exist, optimised for different tasks. Simple feedforward networks pass information solely from input to output and suit basic classification tasks. Convolutional networks are specialised in processing images, while recurrent networks were developed for sequential data such as speech.
With the Transformer architecture a particularly capable variant emerged that today underpins large language models. What all variants share is the basic principle of connected neurons and learnable weights. Deep neural networks with many hidden layers also form the foundation of Deep learning.
What are neural networks used for?
Artificial neural networks are used wherever patterns are to be recognised or predictions derived from complex data. Typical fields are image and speech recognition, machine translation, recommendation systems and detecting fraud or anomalies.
Generative systems producing text, images or music rest on neural networks too. Their great strength is the ability to learn from examples and generalise to new, unfamiliar situations. Tasks can thus be automated that classic rule-based programs could hardly solve.
Using neural networks sensibly with Elisabit
Neural networks are a powerful tool but only show their value when matched to the right use case. Elisabit helps companies judge the potential of neural networks and AI models realistically and turn it into workable solutions without being led by buzzwords.
We concentrate on using modern AI technologies where they create measurable value and integrate them seamlessly into existing processes and websites. An abstract technology thus becomes a concrete contribution to value creation.
Frequently asked questions
Does a neural network really work like the human brain?
Only to a very limited extent. Artificial neural networks are inspired by the brain but heavily simplified and mathematically defined. They reproduce neither the biological complexity of real nerve cells nor their workings in full. The analogy helps understanding but should not be taken literally.
What is the difference between a neural network and deep learning?
Deep learning means training neural networks with a particularly large number of hidden layers, so-called deep networks. A neural network is the basic building block, while deep learning describes the method of using such deep networks for complex tasks.
What do the weights in a neural network mean?
The weights determine how strongly the signal between two neurons is passed on. They are a network's actual learnable parameters. During training they are adjusted so the network produces the desired outputs as accurately as possible.
How many layers does a neural network need?
That depends on the task. Simple problems can often be solved with few layers, while complex tasks such as language processing need very deep networks with many layers. More layers raise capability but also the need for data and computing power.
Related terms
Deep learning uses deep neural networks to recognise complex patterns in large volumes of data automatically.
Machine learning enables systems to learn from data and make predictions without being explicitly programmed.
The transformer is an AI architecture that captures relationships within text using the attention mechanism.
Generative AI independently creates new content — text, images, audio or code — based on patterns it has learned.
An LLM is an AI language model that understands and produces text by predicting the most likely next word.
Put AI to work for your business?
We help you integrate artificial intelligence into your processes, your marketing and your website — strategically and securely.
Your contact
Stefan
I look forward to hearing about your project and finding the best solution together.
