Machine learning encompasses several different types of AI. And there are multiple types of machine learning algorithms that individuals or organizations use, which depend on their needs. Here are the three major categories you should know about for a basic understanding of machine learning technology.
Supervised learning is a type of machine learning in which a human acts as a teacher to the computer. Through supervised learning, a computer processes some training data and the responses that should come from that data— an input and output. The algorithm is then able to analyze both the input and the output, determine how they’re related, and come up with a function.
You can think of supervised learning as akin to how humans learn. You go to class with a teacher who instructs you with example problems and their solutions. Then you’re able to come up with a sort of rule or formula to use, which you can then apply to different scenarios as you encounter them.
One of the most common places to encounter supervised learning is in recommendation engines, which you see on entertainment platforms like Hulu, Netflix, Spotify, or Pandora. These recommendation engines look at the options people have to watch and what they choose to watch, then offer up suggestions based on that behavior.
Unsupervised machine learning is a bit more sophisticated than supervised machine learning; it does not require significant human intervention. In unsupervised learning, the computer takes an algorithm made of unlabeled data and learns from it without needing to be given any responses.
The computers that can do unsupervised learning are incredibly powerful. Because of this, they can predict patterns in the data that they analyze and derive meaning from those patterns. This is particularly relevant for large datasets, which are difficult for humans to analyze and identify patterns in. Many researchers are also employing unsupervised learning to discover rules and group data points.
Today firms are deploying this technology in a number of contexts; one standout: marketing and advertising. Using unsupervised machine learning, an advertising platform can differentiate a large group of consumers into smaller groups based on characteristics like age or income and develop more targeted campaigns. This is one of the tools that has allowed marketing to get more precise, targeted, and effective over the last few years.
The final type of machine learning that you should know about is reinforcement learning. In reinforcement learning, a computer is provided algorithms (called agents) without their solutions. The computer is tasked with coming up with a solution to these algorithms for which it can be given feedback. Through this feedback it learns and improves its accuracy.
You can think of this approach as the machine learning equivalent of learning by trial and error. When the computer gets something right, it knows that it solved something the right way. When it’s told that it’s come up with the wrong solution, it knows to approach the algorithm differently.
The field of chatbots has in recent years extensively employed reinforcement learning. Dialogue systems look at previous customer interactions, then use the results and feedback of those interaction to improve over time. Ultimately, each chatbot-human interaction should be better than the last, because the chatbot takes the reinforcement it was (or wasn’t) given and applies it to future interactions. Experts are also introducing reinforcement learning into the tutoring world, since it will allow companies to create personalized learning experiences.
Each type of machine learning requires different levels of human intervention, and each makes sense to use in specific situations. If you want to make the best usage of machine learning possible, it’s important to understand each type. Then, you can apply the type that works best for your scenario.