Deep Learning
Deep learning is a branch of machine learning that uses neural networks with many layers to learn complex patterns from large amounts of data, powering most of today's advanced AI systems.
What is Deep Learning?
Machine learning gave computers the ability to learn from data rather than follow hand-written rules. Deep learning took that idea and pushed it much further. The 'deep' in deep learning refers to the number of layers in the neural network doing the learning — instead of one or two, a deep learning model might have dozens or even hundreds of layers, each one extracting progressively more abstract patterns from the data passing through it.
The first layers of a system trained to recognize images might detect edges and basic shapes. Deeper layers combine those into textures and objects. The deepest layers put it all together and identify what is actually in the picture. Deep learning largely reduces the need for manual feature engineering. Earlier machine learning approaches required engineers to manually decide which aspects of the data the model should pay attention to — a time-consuming process that also capped how good the system could get. Given enough data, computing power, and the right architecture, a deep learning model works out the relevant features on its own, often identifying patterns that human engineers would never have thought to look for.
Most modern advanced AI capabilities that have captured public attention in recent years run on deep learning. The large language models behind ChatGPT and Claude are deep learning systems. So are the image generators that turn a sentence into a picture, the models that predict how proteins fold to speed up drug discovery, and the tools that restore a blurry old photograph to sharp detail. Deep learning did not invent artificial intelligence, but it is the reason AI went from a promising research field to a technology reshaping entire industries.
Real-world example
When you use Google Photos to search for 'beach' and it pulls up every holiday photo you have ever taken near the ocean — without you ever tagging a single image — that is deep learning at work. The system was trained on millions of labeled images and learned to identify sand, water, and coastal scenes well enough to find them in your personal library.
Related terms
Frequently asked questions
What is the difference between machine learning and deep learning?
Deep learning is one approach within the broader field of machine learning — every deep learning system is a machine learning system, but plenty of machine learning uses simpler methods instead. The practical dividing line is feature engineering: traditional machine learning usually needs humans to decide which aspects of the data matter, while deep learning works those out for itself using many-layered neural networks. That is why it pulls ahead on complex tasks like image recognition and language understanding.
Why does deep learning require vast data and computing power?
Because it is starting with no task-specific knowledge. A deep learning model begins with randomly set internal values and adjusts them incrementally based on the examples it sees. The more layers it has, the more examples it needs to tune all those values reliably. A model with many parameters — the internal values being adjusted — needs a substantial volume of examples to learn from, otherwise it either fails to generalize or memorizes the training data without learning meaningful patterns. The exact amount varies widely depending on the task, the architecture, and whether the model is building on prior training.
What can deep learning do that older machine learning could not?
The big shift is that deep learning can work directly with raw, messy data — images, audio, natural language — without a human first hand-crafting the features for it to look at. Older machine learning could do impressive things with neatly structured data like spreadsheets and tables, but it struggled with a photograph or a paragraph of free text. Deep learning's knack for learning straight from raw input is what unlocked the wave of capabilities — fluent chatbots, realistic image generation, and much more — that most people now associate with AI.