In contrast to Zero-Shot Learning, few-shot learning is a type of ML where a model is trained to generalize and make accurate predictions based on very few examples per class. It is useful when a large amount of labelled data is impractical or expensive.
There are a few main approaches to achieving few-shot learning:
- Metric-Based Learning: learn a similarity measure between examples so the model can compare new data points to a small set of labelled examples
- Optimization-Based Learning: adapt the model parameters efficiently using a small amount of task-specific learning
- Transfer Learning: finetune a model pre-trained on a large dataset with a few available examples for the new task
- Data Augmentation: synthetic data generation or augmentation to expand training data for better learning
For example, if you want a model to classify new animal species and you only have 3 labelled images per species, few-shot learning enables the model to perform this classification effectively by leveraging patterns learned from previous tasks. It is quite similar to how humans learn.