Demystifying Machine Learning

Summary:

• Three main machine learning settings
  • Supervised learning: Ground truth labels given at training time
  • Unsupervised learning: Ground truth labels not given. Aim to discover hidden structure in data.
  • Reinforcement learning: The input and feedback signals come from the live environment
• Other variants
  • Semi-supervised learning: Only some ground truth labels are given
  • Weakly-supervised learning: Ground truth labels are not as detailed
  • Few/zero-shot learning: Few/no examples for labels are given
• Always understand and examine your annotation labels!
  • Imbalanced dataset