What Is Deep Learning?
In the previous lesson, we learned that machine learning allows computers to learn useful patterns from data.
Deep learning is a specialized approach to machine learning that uses neural networks with multiple layers to learn increasingly complex patterns from data.
You can think of deep learning as one level deeper in our journey:
Artificial Intelligence → Machine Learning → Deep Learning
AI is the broad field. Machine learning is one major approach used to create AI systems. Deep learning is a powerful type of machine learning based on neural networks with multiple layers.
What Is a Neural Network?
A neural network is a mathematical model inspired loosely by the way networks of neurons are organized in the human brain.
The word "inspired" is important. An artificial neural network is not a digital copy of a human brain. It is a mathematical system designed to process information and learn patterns.
A neural network contains connected units that are commonly organized into layers.
At a very simplified level, we can imagine:
- An input layer that receives information.
- One or more hidden layers that process the information.
- An output layer that produces a result.
Why Is It Called Deep Learning?
The word deep refers to the presence of multiple layers between the input and the output.
A simple neural network might contain only a small number of layers. A deep neural network contains many layers that can learn different levels of representation.
For example, when processing an image, earlier layers might learn to recognize simple patterns such as edges and shapes. Later layers can combine those patterns into more complex structures. Eventually, the network can use those representations to help identify objects or other features in the image.
The important idea is that the system can learn increasingly complex representations as information moves through the layers.
A Simple Image Example
Imagine training a deep learning system to recognize cats in photographs.
A photograph contains a huge number of individual pixels. Trying to write rules manually such as "a cat has two ears, two eyes and a particular nose shape" would not work reliably across different photographs.
A deep learning model can instead learn patterns from many examples of images.
At a simplified conceptual level:
- Early layers can learn simple visual patterns.
- Middle layers can combine those patterns into shapes and parts.
- Later layers can combine those features to recognize more complex objects.
The model is not being given a complete set of rules describing what a cat looks like. It learns useful representations from the training examples.
Deep Learning Still Learns From Data
Deep learning does not remove the need for data. In fact, data is one of the major ingredients that makes modern deep learning systems possible.
During training, the neural network processes examples and adjusts its internal parameters so that its predictions become more accurate.
For example, a system trained to recognize handwritten numbers might receive thousands or millions of examples of handwritten digits. Over time, the model adjusts itself so that it becomes better at recognizing the patterns associated with each digit.
Why Did Deep Learning Become So Powerful?
Deep learning existed as an area of research for many years, but its capabilities grew dramatically as several important technologies developed together.
Three particularly important factors are:
- More data: The digital world produces enormous amounts of useful training data.
- More computing power: Modern hardware can perform the large number of mathematical calculations required to train neural networks.
- Better algorithms and techniques: Researchers developed improved methods for training and using neural networks.
The combination of these factors helped deep learning achieve major advances in areas such as computer vision, speech recognition and natural language processing.
Where Is Deep Learning Used?
Deep learning is now used in many applications.
- Image and object recognition
- Speech recognition
- Language translation
- Natural language processing
- Recommendation systems
- Fraud and anomaly detection
- Medical image analysis
- Autonomous systems
- Generative AI
Many modern AI products rely on deep learning somewhere in their technology stack.
Machine Learning vs Deep Learning
Deep learning is part of machine learning, so the two terms should not be treated as completely separate technologies.
A useful way to remember the relationship is:
AI is the broad field.
Machine learning is a major approach within AI that allows systems to learn patterns from data.
Deep learning is a type of machine learning that uses multi-layer neural networks to learn complex representations.
What Makes Deep Learning Different?
One important advantage of deep learning is its ability to learn useful representations directly from large amounts of relatively raw data.
In some traditional machine learning systems, people may need to carefully design or select useful features before training the model.
Deep learning can often learn many of those representations automatically through its layers.
This does not mean that humans are no longer needed. People still choose the problem, collect and prepare data, design experiments, select models, evaluate results, and decide how the system should be used.
Deep Learning Has Limitations
Deep learning is powerful, but it is not magic.
Training large neural networks can require substantial amounts of data, computing resources, time, and expertise.
Deep learning models can also make incorrect predictions. A model can learn unwanted patterns from its training data, and its performance can change when it encounters data that is very different from what it saw during training.
Another challenge is that some deep learning systems can be difficult to interpret. It may not always be easy to explain exactly why a complex model produced a particular prediction.
Deep Learning and Generative AI
Deep learning is also one of the foundations of modern generative AI.
Generative AI systems can produce new content such as text, images, audio, video, or computer code.
Many of the modern systems behind these capabilities use large neural networks trained on very large datasets.
We will explore generative AI in much greater detail later in this course.
The Big Picture
Deep learning is a powerful form of machine learning based on neural networks with multiple layers.
Its ability to learn complex representations from large amounts of data has made it an important technology behind many modern AI systems.
The key progression to remember is:
AI → Machine Learning → Deep Learning
As we continue, we will build on this foundation and look at generative AI, large language models, AI assistants, and the technologies behind the tools people are increasingly using every day.