Deep learning is a type of machine learning that uses neural networks with many layers. The word "deep" refers to this large number of layers. Each layer turns simple features into more complex ones.
A visual example#
In recognizing a cat photo, the first layers see only edges and colors, the middle layers detect ears and eyes, and the last layers conclude that the whole image is a cat. In traditional machine learning, people had to define these features by hand; in deep learning, the model finds them itself.
Why it suddenly became important#
Three things came together: the huge amount of data on the internet, powerful graphics cards that made parallel computation possible, and better algorithms. That is why the big leaps in image recognition, machine translation, speech recognition and language models happened in roughly the last decade.
Where it is used#
- Translation and speech recognition
- Face recognition and medical imaging
- Self-driving cars
- Chatbots and image generation
Limitations#
Deep learning usually needs a lot of data and computing power and often acts like a black box; that is, it is hard to explain simply why the model made a given decision.
Common mistake#
Not every problem needs deep learning. For small tabular data, simpler methods are often faster, cheaper and more explainable.