Semi-Supervised Learning Method

A Semi-Supervised Learning Method is a machine learning method that combines labeled and unlabeled data to improve model performance beyond what either data type alone would achieve.

• AKA: SSL Method, Hybrid Learning Method, Partially Supervised Method.
• Context:
  • It can typically leverage Unlabeled Data through semi-supervised assumptions.
  • It can typically reduce Annotation Cost via semi-supervised label efficiencys.
  • It can often improve Model Generalization through semi-supervised data utilizations.
  • It can often enable Practical Deployment via semi-supervised resource optimizations.
  • It can range from being a Generative Semi-Supervised Method to being a Discriminative Semi-Supervised Method, depending on its semi-supervised modeling approach.
  • It can range from being a Graph-Based Semi-Supervised Method to being a Assumption-Based Semi-Supervised Method, depending on its semi-supervised theoretical foundation.
  • It can range from being a Transductive Semi-Supervised Method to being a Inductive Semi-Supervised Method, depending on its semi-supervised inference scope.
  • It can range from being a Single-View Semi-Supervised Method to being a Multi-View Semi-Supervised Method, depending on its semi-supervised data perspective.
  • ...
• Examples:
  • Self-Learning Semi-Supervised Methods, such as:
    • Self-Training Algorithm for confidence-based pseudo-labeling.
    • Pseudo-Label Method for hard label assignment.
    • Entropy Minimization Method for prediction sharpening.
  • Co-Learning Semi-Supervised Methods, such as:
    • Co-Training Algorithm for multi-view agreement.
    • Tri-Training Method for disagreement-based learning.
    • Democratic Co-Learning for consensus building.
  • Consistency Semi-Supervised Methods, such as:
    • Temporal Ensembling for prediction stability.
    • Mean Teacher Method for model averaging.
    • Virtual Adversarial Training for smoothness regularization.
  • Graph Semi-Supervised Methods, such as:
    • Label Propagation for similarity-based spreading.
    • Graph Convolutional Method for structural learning.
    • Manifold Regularization for geometric consistency.
  • ...
• Counter-Examples:
  • Supervised Learning Method, which uses only labeled data.
  • Unsupervised Learning Method, which uses no labels at all.
  • Active Learning Method, which queries for specific labels rather than using existing unlabeled data.
• See: Machine Learning Method, Semi-Supervised Learning, Unlabeled Data, Pseudo-Labeling, Label Propagation, Consistency Regularization, Training Strategy.

---