AI.info
Kinds of learning
Supervised, unsupervised, self-supervised: the feedback signal decides everything downstream, including what can go wrong.
- Learning Approaches: Start With the Feedback
- Supervised Learning: Learning From Labeled Examples
- Labels, Targets, and the Limits of Ground Truth
- Classification: Categories, Scores, and Decisions
- Regression: Predicting Quantities Without Pretending to Know the Future
- Ranking, Scoring, and Structured Prediction
- Learning From Preferences and Pairwise Comparisons
- Unsupervised Learning: Searching for Structure Without Task Labels
- Clustering and Segmentation: Useful Groups Without Invented Essences
- Dimensionality Reduction and Representation Discovery
- Anomaly, Novelty, and Density: Finding What Does Not Fit
- Generative and Discriminative Modeling
- Reinforcement Learning: Learning From Actions and Consequences
- States, Actions, Rewards, and Policies
- Exploration, Exploitation, and Contextual Bandits
- Imitation Learning: Learning Behavior From Demonstrations
- Semi-Supervised Learning: A Few Labels, Many Unlabeled Examples
- Self-Supervised Learning: Creating Training Signals From the Data Itself
- Weak Supervision: Combining Noisy Sources Instead of Waiting for Perfect Labels
- Active Learning and Human-in-the-Loop Systems
- Transfer Learning and Domain Adaptation
- Batch, Online, and Continual Learning
- Prediction, Causation, and Counterfactual Questions
- Multi-Task and Hybrid Learning Systems
- Choosing the Learning Strategy Under Real Constraints
- Capstone: Design a Learning System and Defend Every Signal