Training & Learning
Unsupervised Learning
Machine learning techniques that find patterns in unlabeled data without explicit target outputs.
Unsupervised learning discovers structure in data without any labels. Instead of learning a mapping from inputs to outputs, it learns the underlying distribution or grouping of the data itself.
Common tasks include clustering (grouping similar items), dimensionality reduction (compressing high-dimensional data), and density estimation. Techniques like k-means, PCA, and t-SNE are classic examples.
Why it matters: labeled data is expensive. Unsupervised methods let you learn from the massive amounts of unlabeled data that exist in the real world.
Modern LLMs rely heavily on self-supervised learning, a variant where the model generates its own labels from raw text by predicting masked or next tokens.