On this page
Deep learning is a specialized subset of machine learning, which is itself a subset of artificial intelligence. The main difference is that deep learning uses multi-layered artificial neural networks that learn useful features from data on their own, while classic machine learning often depends on people choosing those features.
Key takeaways
- Artificial intelligence (AI) is the broad field, machine learning sits inside it, and deep learning sits inside machine learning.
- Deep learning uses layered artificial neural networks inspired by the human brain.
- Deep learning needs less manual feature selection but usually more data and more training time.
- Both approaches have strengths, so neither is always the better choice.
How are AI, machine learning, and deep learning related?
The term artificial intelligence was coined by John McCarthy in 1956. In simple terms, AI is the development of computer systems that can perform tasks that normally require human intelligence.
Machine learning and deep learning both grew out of AI. Deep learning is a subset of machine learning, and machine learning is a subset of AI. Strictly speaking, then, asking about deep learning versus machine learning is asking what makes one kind of machine learning special.
What is machine learning?
Machine learning is a set of algorithms that parse data and learn from experience and mistakes so they can make informed decisions. These algorithms typically work on structured data, and the learning may or may not be supervised, depending on the data they receive. They do not need to be recoded each time the input changes.
Amazon is a useful example. Its machine learning systems use customer data to decide which offers to send and when, which products to show on a user’s home page, and how to manage stock. Machine learning is also widely used by large companies to improve products, automate tasks, and detect malware.
What is deep learning?
Deep learning is a more sophisticated and mathematically complex form of machine learning. It uses artificial neural networks, which are layered algorithms inspired by the structure and function of the human brain. The term is commonly traced to Igor Aizenberg, who used it in 2000.
The field grew in importance as the internet produced huge volumes of data that had to be sorted for products and customer benefit. Deep learning copes well with large amounts of data, though it generally takes longer to train.
A practical example: recognizing a stop sign
Driver-assistance features in cars such as Tesla vehicles are often described as using deep learning to recognize objects like stop signs, nearby vehicles, and road dividers. Take the stop sign as an example.
With traditional machine learning, a software engineer must choose the relevant features, such as edges, colors, shapes, and size, and feed many example images to the algorithm. When the sign looks different, features may need to be reselected.
With deep learning, the neural network identifies features such as edges, colors, and size by itself and keeps learning from its own errors as it sees more training data.
What are the main differences between machine learning and deep learning?
| Aspect | Machine learning | Deep learning |
|---|---|---|
| Structure | Simpler algorithms | Multi-layered neural networks |
| Training time | Generally shorter | Generally longer |
| Human intervention | Often needs manual feature selection | Learns features largely on its own |
| Data handling | Works well with structured data | Handles large volumes of structured and unstructured data |
| Decision speed | Depends on the model | A trained network can make decisions quickly |
Training
Because machine learning models are simpler, they usually take less time to train. Deep learning models rely on complex, interconnected networks, so they typically need considerably more training time and computing power.
Human intervention
Deep learning needs little manual input once it is trained. For example, a network can learn to recognize faces from features like eyes, nose, and ears without extra labeled examples for every different look.
Data handling
Deep learning is better suited to very large datasets, and to data that is unstructured. That does not make classic machine learning obsolete, since each approach has its own advantages.
Execution time
Once trained, a deep learning model can often reach a decision faster than many traditional approaches. Combined with the growth of data of all types, this is one reason deep learning has become so popular.
Frequently asked questions
Is deep learning the same as machine learning?
Not exactly. Deep learning is a subset of machine learning that uses multi-layered neural networks.
Does deep learning always beat machine learning?
No. Deep learning tends to shine with large datasets and complex inputs, while simpler machine learning methods can be faster and cheaper on smaller problems.
Why does deep learning need so much data?
Neural networks learn their own features, which usually takes many examples to do reliably.
Who coined the term artificial intelligence?
John McCarthy, in 1956.
Conclusion
Deep learning is a powerful form of machine learning that trades heavier data and training needs for less manual feature work. Both techniques will keep shaping technology jobs, and learning them now is a sound investment for developers and programmers.



