Don't stop here
Hand-picked guides our readers explore right after this one.
Stunning image generation with Midjourney prompt mastery
Read the guideExpert guide to Claude prompts with XML tags, artifacts, and complex reasoning
Read the guideAI prompts for product strategy, user research, roadmapping, and stakeholder communication
Read the guideA computing system loosely inspired by the brain, made of connected nodes that pass and transform signals to learn patterns.
A neural network is a system of connected units, called neurons, that pass numbers to each other. Each connection has a weight that strengthens or weakens the signal. By adjusting these weights based on examples, the network learns to turn inputs (like the pixels of an image) into outputs (like the label 'cat'). Stack enough neurons and layers and the network can learn remarkably complex patterns. Neural networks are the foundation of deep learning and modern AI.
Imagine a huge network of water pipes with adjustable valves. Water (data) flows in one end and out the other. By tuning thousands of valves (weights), you can control exactly how input flows to output. Training is the process of adjusting all those valves until the right input produces the right output. The network 'learns' by valve-tuning, not by being told rules.
An artificial neural network is composed of layers of nodes; each node computes a weighted sum of its inputs, adds a bias, and applies a non-linear activation function (such as ReLU). Information flows forward from input to output layers. Learning happens via backpropagation: the network's error is measured by a loss function, and gradients of that loss with respect to each weight are computed and used to update the weights through gradient descent. Non-linear activations are essential, they let networks approximate complex functions that linear models cannot. Depth and width, plus specialized architectures, determine capacity.
The classic first demo: reading handwritten digits from pixels.
Specialized neural networks that excel at images.
The neural network architecture behind LLMs like ChatGPT and Claude.
Neural networks power feed ranking and recommendations across major apps.
Only loosely. They were inspired by biological neurons, but artificial neural networks are simplified mathematical functions, not biological simulations. Real neurons are far more complex, and the brain learns very differently from backpropagation. The brain analogy is a helpful metaphor, not a literal description.
A neural network is the structure; deep learning is the practice of using neural networks with many layers to learn from large data. A shallow neural network with one hidden layer is a neural network but not really 'deep learning'. Deep learning is neural networks scaled up in depth and data.
It makes a prediction, measures how wrong it was against the correct answer (the loss), and then adjusts its internal weights slightly to reduce that error, a process called backpropagation with gradient descent. Repeat this over millions of examples and the weights settle into values that capture the patterns in the data.
A type of machine learning that uses neural networks with many layers to learn complex patterns from large amounts of data.
⚙️The neural network architecture behind modern AI, introduced by Google in 2017 and now powers ChatGPT, Claude, and most other LLMs.
📈A branch of AI where systems learn patterns from data and improve at a task, instead of being explicitly programmed with rules.
📚A neural network trained on massive text data to understand and generate human-like language.
Our free AI course teaches you to use these ideas in real projects.
Start Free AI Course →