Back to glossary

One-shot Learning

A machine learning task where the model is required to learn information about a category from only one training example.

Categoryai

Overview

One-shot learning is common in computer vision, particularly for facial recognition. A system might need to recognize a person after seeing only one photo of them.

Techniques

  • Siamese Networks: Comparing two inputs to see how similar they are.
  • Memory-Augmented Neural Networks: Storing information about new classes in an external memory.

Contrast with Traditional ML

Traditional deep learning often requires thousands of examples per class, making one-shot learning a significant challenge and a highly desirable capability.

Related Terms

Zero-shot LearningFew-shot Learning