The cost of launching products such in speech recognition, internet content classification, drug discovery and protein sequencing can be prohibitive.

The crucial issue is that teaching algorithms requires an enormous and at the time not available amount of data labeling.

In these situations, the technology utilized is semi-supervised learning. The semi-supervised learning general characteristics are described in this article:

https://www.geeksforgeeks.org/ml-semi-supervised-learning/

This approach can subject to more errors than supervised learning and ultimately is only as good as thee data labeling quality at disposal.

For this reason Facebook has made some research in this field to try to obviate to limitations and fully exploit the unlabeled data and produced an algorithms that can make more precise predictions than algorithms trained with supervised methods:

https://towardsdatascience.com/facebook-believes-in-omni-supervised-learning-78f37253f4f4

Facebook is not the only company working on improving semi supervised learning.

A crucial part of semi-supervised learning is unsupervised learning. In this video interview I made last week,  Amethix Chief Scientist Describes why semi-supervised learning is important and what are the frontiers in healthcare , medical imaging, finance and other industries.