Self-supervised learning
AI learning from raw data by creating its own labels, rather than relying on humans to mark examples.
In Plain English
Self-supervised learning flips the usual script: instead of paying people to label thousands of photos as 'cat' or 'dog,' the model teaches itself by finding hidden patterns in the data. For example, it might predict the next word in a sentence by hiding it and trying to guess—or mask part of an image and learn to fill it in. This approach uses the sheer volume of unlabeled data (text, video, photos) available on the internet, making it much cheaper and faster than traditional training. It's why modern AI can work with so little human guidance.
💡Real-World Example
A medical imaging company has millions of unlabeled X-rays in its archives. Instead of hiring radiologists to annotate each one, they use self-supervised learning: the model learns to reconstruct hidden portions of images, building a deep understanding of normal anatomy without any human labels.
Related Terms
What did you think of our explanation?