Is deep learning always the right tool?

No. It shines on rich, unstructured data like images, audio and text. On small or tabular datasets, simpler models are often more accurate, cheaper and easier to explain.

Why does it need so much compute?

Many layers and millions or billions of parameters mean enormous numbers of calculations during training — which is why GPUs, and the data to feed them, were the practical unlock.

What is Deep Learning? — Getting Digital glossary

Deep Learning

Also: deep neural networks

Deep learning is machine learning with many-layered neural networks that learn their own features directly from raw data, rather than relying on features a human picks out first.

Let the model find the features

For decades, the hard part of machine learning was feature engineering — a human deciding which measurements mattered before any learning began. Deep learning's breakthrough was to hand that job to the model. Give a deep neural network enough raw data and compute, and its layers discover useful representations on their own, each layer building on the one below.

Different shapes of network suit different data: convolutional networks for images and computer vision, and Transformers for language and sequences. The same family powers today's generative AI, from image generators to large language models.

What changed to make it work

The ideas are decades old; what unlocked them this century was the combination of large datasets, fast GPUs, and better training tricks — not a single new equation.

Key takeaways

→Deep learning replaces hand-picked features with representations the network learns itself.
→Its progress came from data and compute meeting old ideas, enabling vision, language and generative models.

Related concepts

Broader topicMachine Learning
Deep learning is a sub-field of machine learning.
IncludesNeural Networks
Neural networks are the building blocks deep learning is made of.
More specificGenerative AI
Generative AI is built on deep-learning models.
More specific

FAQ

Is deep learning always the right tool?: No. It shines on rich, unstructured data like images, audio and text. On small or tabular datasets, simpler models are often more accurate, cheaper and easier to explain.
Why does it need so much compute?: Many layers and millions or billions of parameters mean enormous numbers of calculations during training — which is why GPUs, and the data to feed them, were the practical unlock.

Deep Learning

Let the model find the features

Key takeaways

Related concepts

Courses that teach this

FAQ