02/ The difference between AI, Machine Learning and Deep Learning
In general, Deep learning is a particular kind of machine learning, and machine learning itself is a subset of AI. In other words, deep learning is machine learning, which in turn is part of the broader field of AI.
Let's dig deeper now so you can understand the core differences between these three fields.
AI: think like a human
AI is a broad branch of computer science concerned with building machines capable of simulating human intelligence. AI includes a lot of different techniques that range from coding explicit rules to more advanced algorithms like deep learning.
AI is usually divided into two categories:
Narrow AI: Narrow AI is a kind of AI focused on performing a single task extremely well. Although these machines seem to be intelligent, they operate in a basic way under a lot of limitations.
Strong AI: Strong AI machines can solve any problem in a human-like way. This is the kind of AI of the robots in the movies. Sometimes you see the term General AI as well, they mean the same thing.
Machine learning: learn from data
Machine learning is a group of AI techniques that mimics the human brain’s capability of learning from experience. Machine learning algorithms allow computers to solve problems using data as examples instead of coding an explicit set of rules, like in traditional software development.
Deep learning: discover patterns
Deep learning is a type of Machine learning that is capable of working with complex, unstructured data like text or images. Deep learning made it possible to solve problems that could not be handled with traditional machine learning techniques.
Examples: Deep learning can be used to automate the detection of production defects, to better customer experience with sentiment analysis, or to detect malfunctioning machinery from audio recordings.
03/ What is the difference between deep learning and machine learning?
How are new deep learning models different from traditional machine learning techniques?
Detect complex patterns automatically
Deep learning can detect complex patterns in data, so that deep learning models can work with unstructured data like text and images. Moreover, deep learning can build efficient decision rules that do not have to be designed by hand, like in traditional machine learning.
Requires more data
Deep learning usually requires more data than traditional machine learning techniques. However deep learning algorithms outperform older machine learning techniques when they have lots of examples to learn from.
Computationally more expensive
If you want to create a deep learning model, you need access to powerful computing resources, i.e, GPU acceleration. Machine learning techniques are less computationally expensive and often require less time to prototype and operationalize.
Interpretability can be an issue
Interpretability in machine learning means being able to understand why and how the model makes a choice. Deep learning models can select independently what features are relevant when solving your problem. This is very useful when finding patterns in unstructured data. However this means that it can be difficult to understand a deep learning model's decision process. With machine learning you have to manually decide the features to include in your model. This can be a long or difficult task but this property makes machine learning models easier to interpret.
04/ When to use deep learning
Choose deep learning to solve your problems when:
You are dealing with unstructured data.
Deep learning is the best technique to deal with text and images.
You want to detect patterns automatically.
For example, you don’t want to rely on domain expertise to extract relevant features from your data or interesting patterns can be difficult to find by hand
You have access to a lot of computing resources.
Deep learning is computationally expensive and you need powerful resources to build your models.
05/ When to use machine learning
Use more traditional machine learning techniques when:
You have a small dataset.
Deep learning works best with a lot of data, while you need less data to obtain good performance with machine learning.
Interpretability is important.
You really need to understand how and why your model makes decisions.
You have limited computational resources.
Machine learning algorithms usually require less computational power than deep learning.
06/ How to use deep learning
Deploy an operational AI model. Build an image classification model that can recognize handwritten numbers.
Buy or not? Predict from tabular data. Predict if a customer will buy or not based on earlier customers buying patterns.
Classify text in any language. Build a multilingual text classifier that can recognize many different topics.