Self-supervised learning is a term used in the fields of artificial intelligence, big data, smart data and automation. It refers to an innovative method that enables computers to learn independently from large amounts of data without the need for many people to prepare or label data manually.
With self-supervised learning, machines try to filter existing information from unstructured data and recognise patterns - similar to humans, who often learn new things by observing them. The special thing about it: The computer sets itself tasks, which it then tries to solve using the available data.
A simple example: Imagine you want a computer to understand how sentences are structured. To do this, it is given many texts in which a word is sometimes omitted. Its task is to guess the missing word. Through many such puzzles, the computer recognises connections in the texts without a human having to comment on each sentence individually.
Self-supervised learning makes it easier for companies to use their own databases and make automated systems more intelligent - from voice assistants to automated quality controls in industry. This results in powerful AI applications, even if no elaborately labelled training data is available.
Deutsch 
English (UK)