Self-supervised learning is the umbrella for methods that generate their own training signal from unlabeled data — predicting the next word in text being the canonical example. Yann LeCun has been the loudest voice arguing this paradigm should sit at the center of AI. The Pre-training phase of modern LLMs is almost entirely self-supervised, which is what makes internet-scale data usable in the first place. It effectively bridges Supervised Learning and Unsupervised Learning in practice.
MEVZU N°124ISTANBULYEAR I — VOL. III
Glossary · Intermediate · 2018
Self-Supervised Learning
A machine learning paradigm in which the model generates its own training signal from unlabeled data.
- EN — English term
- Self-Supervised Learning
- TR — Turkish term
- Öz-Denetimli Öğrenme