Few-Shot Learning

Few-Shot Learning techniques are used to dealing with minimal amount of training data for training Machine Learning models.


Low quantity of data may be caused by low availability of data or annotation cost.


  • Data-Level Approaches: Solving data quantity problem via Data Mining.
  • Parameter-Level Approaches: Generalize the model by using Regularization and Loss Functions methods, or alternatively use very large parameter space to optimize results.
  • Meta learning or it's variations such as Model-Agnostic Meta-learning.
  • Metrics-Based Approaches