Few-Shot Learning

Few-Shot Learning (FSL): A meta-learning paradigm that enables models to generalize new concepts using only a handful of labeled examples (N-way K-shot) by leveraging prior knowledge.

FSL is a crucial machine learning framework designed to combat data scarcity: it trains a model to 'learn how to learn' from a distribution of tasks, not just a single dataset. This meta-learning approach allows for rapid adaptation to new classification problems using minimal data points—typically 1 to 5 per class. Key algorithms, including Model-Agnostic Meta-Learning (MAML) and Prototypical Networks, drive this capability. FSL delivers immediate value in high-cost or rare-data domains, such as medical diagnosis (e.g., classifying rare diseases) or detecting newly identified species, significantly reducing the dependency on massive, expensive labeled datasets.

https://www.ibm.com/topics/few-shot-learning