What is Few-Shot Learning?

Few-shot learning is a machine learning approach where models can learn new tasks or concepts from only a small number of examples (typically 1-10). It contrasts with traditional ML that requires large datasets and leverages knowledge from pre-training.

Learning Paradigms

Zero-Shot Learning No examples provided. Uses task description only.

One-Shot Learning Single example provided. Common in image recognition.

Few-Shot Learning 2-10 examples provided. Balance of guidance and efficiency.

In Large Language Models

Few-Shot Prompting

Translate English to French:
cat → chat
dog → chien
house → ?

Techniques

Meta-Learning

Learn to learn
MAML, Prototypical Networks
Task-agnostic representations

Transfer Learning

Pre-trained models
Fine-tuning on few examples
Domain adaptation

In-Context Learning

Examples in prompt
No weight updates
LLM capability

Applications

Personalization
Rare class classification
Drug discovery
Robotics
Language tasks

Challenges

Example selection importance
Task ambiguity
Overfitting to few examples
Evaluation difficulty

Best Practices

Choose representative examples
Use diverse examples
Order matters in prompts
Test with different examples
Combine with clear instructions