Neural Network
A neural network is an AI model made of layered, interconnected nodes loosely inspired by neurons in the brain, which learns patterns from data by adjusting the strength of connections between those nodes.
Key takeaways
- A neural network is made of layered nodes connected by weights, loosely inspired by neurons in the brain.
- Networks learn by adjusting their weights to reduce prediction error, a process called backpropagation.
- Different neural network architectures suit different data types — convolutional networks for images, transformers for language.
- Neural networks learn patterns directly from data rather than following hand-written rules.
- Nearly all modern AI systems, including large language models, are built on neural networks.
What is a neural network?
A neural network is a computing structure made up of layers of simple units called nodes, or neurons, each connected to nodes in the next layer. Data enters through an input layer, passes through one or more hidden layers that transform it, and produces a result at an output layer. The strength of each connection, called a weight, determines how much influence one node has on the next.
How neural networks learn
A network starts with random weights and gradually adjusts them through training: it makes a prediction, compares that prediction to the correct answer, and nudges its weights to reduce the error, repeating this millions or billions of times across a training dataset. This process, called backpropagation combined with gradient descent, is how a network goes from producing random output to producing accurate predictions.
Types of neural networks
Different architectures suit different problems. Convolutional neural networks are effective at recognizing patterns in images. Recurrent networks were historically used for sequences like text or audio. The transformer architecture, which underlies large language models, is now the dominant design for language and increasingly for other data types too.
Why neural networks matter
Neural networks are the foundation nearly all modern AI is built on, from image recognition and speech-to-text to the large language models behind today's chatbots. Their ability to learn patterns directly from data, rather than following hand-written rules, is what makes them adaptable to so many different problems.
Frequently asked
- What is a neural network in simple terms?
- A neural network is an AI system made of connected layers of simple units that learns to make predictions by adjusting the strength of those connections based on training data.
- Is a neural network the same as a large language model?
- No. A neural network is the general concept; a large language model is a specific, very large neural network (usually a transformer) trained on text to understand and generate language.
- How does a neural network learn?
- It compares its predictions to correct answers, calculates the error, and adjusts its internal weights through backpropagation to reduce that error over many training examples.