AI Trend Research

← AI glossary

AI glossary · Beginner · Language models

Deep learning

Deep learning is a machine learning method that stacks many layers of artificial neural networks, modeled on the neural networks of the human brain, so that the model learns complex patterns from data on its own.

About this term
Level
Beginner · Core ideas for first-time readers
Category
Basics
Added
2026-10-08
In plain words Like the workstations on a factory line, the first layer looks at simple features such as lines or dots, and later layers combine them to recognize eyes and noses and, further on, complex concepts such as 'a face'. It does not actually work the same way as a real brain; it just borrows the idea from the brain's structure.

How it works

A deep learning model is an artificial neural network made up of an input layer, several hidden layers and an output layer. Each layer computes on the numbers it receives from the previous layer and passes the result to the next, capturing more abstract features as it goes deeper. During training, a huge number of parameters are adjusted using backpropagation, which passes the difference between the output and the correct answer backward, and gradient descent, which reduces the error little by little. The 'deep' refers to these many layers.

A layered deep learning modelInput layer → Hidden layer 1; Hidden layer 1 → Hidden layer 2; Hidden layer 2 → Output layer; Output layer → Backpropagation(Error)Input layerPhoto pixelsHidden layer 1Lines/dots (example)Hidden layer 2Eyes/nose (example)Output layer'Face' decisionBackpropagationPasses the error backwardError
A layered deep learning model As input passes through several hidden layers, it changes from simple features into complex concepts. The features written on the layers are examples assuming image recognition.

Why it matters

Earlier machine learning often required people to choose in advance 'which features to look at'. Deep learning finds the features it needs on its own from raw data such as the pixels of a photo or the characters of a sentence. This greatly improved performance in areas such as image recognition, speech recognition and translation. Today's generative AI, including LLMs and image generation AI, is all built on deep learning.

Things to know

Deep learning requires a lot of data and powerful computing resources such as GPUs. Because it is hard to tell why the model produced a given answer, it is criticized as a 'black box', and the research that tries to address this is interpretability. Well-known architectures include CNNs, which are strong on images; RNNs, used for sequential data; and the transformer, the basis of today's LLMs.

Example

A smartphone photo app automatically sorting photos by person, a voice assistant turning speech into text, and a chatbot understanding sentences and replying all run on deep learning models.

Related terms

Machine learning Artificial neural network Backpropagation Transformer CNN GPU