Självövervakad

Självövervakad is a Swedish word that translates to "self-supervised" in English. It refers to a type of machine learning where a model learns from data that has not been explicitly labeled. Instead of relying on human annotators to provide the correct answers, self-supervised learning generates its own supervisory signals from the input data itself.

This approach is particularly useful in domains where obtaining large amounts of labeled data is expensive or time-consuming, such as natural language processing or computer vision. The model is typically trained on a "pretext task" where it has to predict a missing or corrupted part of the data. For example, in image recognition, a model might be tasked with predicting the rotation of an image or filling in a masked-out patch. By solving these pretext tasks, the model learns general representations and features of the data that can then be transferred to downstream tasks with less labeled data.

Self-supervised learning has become a prominent area of research in artificial intelligence, leading to significant advancements in areas like language modeling and image understanding. It allows models to leverage the vast amounts of unlabeled data available on the internet and in various datasets, enabling them to learn more robust and generalizable patterns without direct human supervision for every training example. This reduces the dependency on expensive manual labeling processes and opens up new possibilities for AI development.