Self-Supervised Learning
Imagine that you are in the middle of the ocean, and you are thirsty. There is water all around you, but you cannot drink any of it. But what if you had the resources to boil the salt out of the water and thereby make it drinkable? Of course, the energy costs associated with the process can be quite high, so you will likely use the process in moderation. However, if your energy costs effectively became free, for example, if you were harnessing the power of the sun, the process might be more attractive for you to do on a larger scale.
In our somewhat simplistic situation described above, the first scenario is roughly analogous to supervised learning, and the second to the class of unsupervised / semi-supervised learning techniques we will cover in this chapter. The biggest problem with supervised learning techniques is the time and expense associated with the collection of labeled training data. As a result, labeled datasets are often relatively small.
...
