Semi-supervised Learning


Semi-supervised learning is a machine learning technique in which some of the data are given labels, while others have no labels, in an attempt to improve the performance of a model with zero data labels.

  • e.g., want to train a model how to identify fractured ceramics for quality assurance
    • feed the model millions of product images
    • but only label some of the data as fractured
    • hoping it can learn how to find fractured ceramics from the small set provided
  • more cost-effective method than supervised learning
    • labeling data can be an arduous task