Supervised and Unsupervised Learning
Definition
In machine learning, the case where the dependent variable is specified is called supervised learning, and the case where it is not specified is called unsupervised learning.
Example
The difference between supervised and unsupervised learning can be simply compared to the difference between multiple-choice and essay questions. For example, let’s say there is a classification problem asking for the color of 6 tiles like above.
Supervised Learning
But here, if there are only two choices, green or red, honestly, some might be half-half and some might even be yellow, but one would choose whatever they think is closer. Thinking a bit differently, it can be seen as having provided the information that the answer is ‘green’ or ‘red’. In this sense, the term supervised learning is appropriate. To generalize a bit more, it is said that ’the dependent variable is given’.
Supervised learning is used when there is interest in regression and classification.
Unsupervised Learning
On the other hand, if asked to distinguish without any restrictions, ‘red-green gradient’ or ‘yellow’ are something clearly distinct from green or red. And that too is a valid answer. In this way, although learning does occur, since the answer comes without any particular restrictions, the term unsupervised learning is appropriate. In this case, the machine does not dwell on whether the answer is green or red. It is up to humans to assign meaning to the answer.
Unsupervised learning is used when there is interest in clustering.
Reinforcement Learning
If there’s something common between supervised and unsupervised learning, it would be that both are methods to solve problems that have answers, whether they are multiple-choice or essay-type questions. Simply put, it’s about responding to the questions asked, but reinforcement learning focuses on ’learning to do well’. Questions like ‘What are we supposed to do well?’, ‘Isn’t it too vague and ambiguous?’ are valid. However, conversely, if there’s ‘something that needs to be done well but it’s vague and ambiguous’, one might feel the need for an approach like reinforcement learning.