Michele LaurelliBlog

Few-Shot Learning

/fjuː ʃɒt ˈlɜːrnɪŋ/

A model's ability to learn from a small number of examples, typically 1-10 examples per class.

Few-shot learning leverages pre-trained knowledge and meta-learning. LLMs demonstrate this through in-context learning with examples in the prompt.

Examples

  • GPT-3 with 5 examples in prompt
  • One-shot image classification
  • Meta-learning algorithms