AI/ML Foundations

Supervised, Unsupervised & Reinforcement Learning

The major learning paradigms differ by feedback signal: supervised learning has labeled answers, unsupervised learning discovers structure, reinforcement learning learns from rewards, and self-supervised learning creates training targets from raw data itself.
  • Supervised learning: examples come with targets
  • Unsupervised learning: there is no target label
  • Reinforcement learning: actions are judged by rewards over time
  • Self-supervised learning: labels are manufactured from the data
  • Feedback signal determines the evaluation plan
Paradigms by feedback signal
ParadigmFeedbackOutputExample
SupervisedKnown labels/targetsPredictorClassify support ticket intent
UnsupervisedNo labelsStructureCluster customers by behavior
Self-supervisedTargets derived from raw dataRepresentation / generatorNext-token pretraining
ReinforcementReward from actionsPolicyGame-playing agent
Sources
  • CS229: Machine LearningCourse description — supervised, unsupervised, reinforcement learning
  • Machine Learning Crash CourseML Concepts
  • Artificial Intelligence: A Modern ApproachLearning