- Purpose: Use the information contained in unlabeled samples to improve the generalization ability of the model
- Active learning: Effectively interact with the outside world to reduce marking costs. Use as few "queries" as possible to get the best possible performance
- Semi-supervised learning: The learner does not rely on external interaction and automatically uses unlabeled samples to improve learning performance
- Pure semi-supervised learning
- Direct learning
- Basic assumption: similar samples have similar outputs
-
- Clustering hypothesis
- Manifold hypothesis
- Four paradigms of semi-supervised learning (classification)
- Generative method
- Semi-supervised SVM
- Figure semi-supervised learning
- Divergence-based approach
- Semi-supervised clustering