Instead of engineers hand-writing rules or features, a deep network learns its own layered representations. Early layers pick up simple patterns and later layers combine them into complex concepts. It is the approach behind modern speech recognition, translation, image generators and chatbots.
How it works
A deep neural network is a chain of layers made of simple units. Each unit multiplies its inputs by weights, adds them up and passes the result through a non-linear function. Stack enough layers and the network can represent very complicated patterns. Training starts with random weights. The network makes a prediction, a loss function measures how wrong it is, and an algorithm called backpropagation works out how each weight contributed to the error. Gradient descent then nudges every weight slightly in the direction that reduces the error. Repeat this over millions or billions of examples and the network improves. Different architectures suit different data: convolutional networks for images, recurrent networks for older sequence work, and transformers, introduced in 2017, for most language and multimodal models today. The field took off around 2012, when a deep network trained on GPUs won the ImageNet image-recognition contest by a wide margin. Large models now have billions of parameters and are trained on clusters of GPUs or similar chips.
Why it matters
Almost every AI tool a creator or business uses today, from auto-captions to background removal to chat assistants, is built on deep learning. Understanding it explains why these tools need so much data and computing power, why they are good at fuzzy, pattern-based tasks and why they can fail in surprising ways. It also helps you judge vendor claims: if a product says it uses AI, it most likely means a deep learning model trained on large datasets.
Where you will see it
- Automatic captions and transcription in video editors and meeting tools.
- Recognising faces, objects and scenes in photos for search and tagging.
- Powering chatbots and writing assistants through large language models.
- Generating images, music and voice from text descriptions.
Limits and common misconceptions
- Needs large amounts of training data and expensive computing power, especially for big models.
- Models are hard to interpret; it is often unclear why a network made a particular decision.
- It learns statistical patterns, not true understanding, so it can fail on inputs unlike its training data.
How it differs from related terms
- Deep learning vs Machine learning Machine learning is the wider field of learning from data; deep learning is the subset that uses neural networks with many layers.
- Deep learning vs Neural network A neural network is the structure itself, while deep learning is the practice of building and training very deep ones at scale.
Frequently asked questions
What is the difference between AI, machine learning and deep learning?
They are nested. Artificial intelligence is the broad goal of machines doing tasks that need intelligence. Machine learning is one way to get there, by learning from data. Deep learning is a type of machine learning that uses many-layered neural networks, and it drives most current AI.
Why is it called deep?
Deep refers to the number of layers between input and output. Older neural networks often had one or two hidden layers. Deep networks have many, sometimes hundreds, which lets them build up complex representations step by step.
Do I need to know maths to use deep learning?
Not to use tools built on it. To build models yourself, it helps to know basic linear algebra, calculus and probability, plus Python. Libraries such as PyTorch and TensorFlow handle most of the heavy maths for you.
Related terms
- Context window The maximum amount of text, measured in tokens, that a language model can take in and work with in a single request, including its reply.
- Copilot An AI assistant embedded inside another app that uses what you are working on as context to suggest, draft or complete tasks.
- Deepfake Synthetic video, image or audio, made with AI, that realistically shows a real person saying or doing something they never said or did.
- Diffusion model A generative model that creates images, video or audio by turning random noise, step by step, into a result that matches the prompt.
- Dubbing (AI) Using AI to translate a video's speech into another language and re-voice it, often in a clone of the original speaker's voice.