📝 Learning Paradigms

Description

< What is it? >

• Purpose: learning paradigms describe how models learn, particularly the data, supervision, or feedback used during training. Examples include supervised learning from labeled examples, self-supervised learning from targets derived from the data itself, and reinforcement learning from rewards obtained through interaction.
• Scope: this category also covers unsupervised learning, which identifies patterns without supplied target labels, and semi-supervised learning, which combines labeled and unlabeled data. A model architecture can be used with different learning paradigms, and a model can undergo several kinds of training.

Related ideas

• Classification and Regression are common supervised prediction tasks.
• Clustering groups similar examples without supplied class labels.
• Reinforcement Learning (RL) learns behavior through rewards.
• Model Families & Architectures describes the models that learning methods can train.

Reference

• Google: Introduction to machine learning
• Google Machine Learning Glossary: Self-supervised learning