Few-shot learning (FSL) is a machine learning paradigm that enables models to generalize to new classes or tasks with a limited number of training examples. This approach is particularly valuable in scenarios where acquiring large labeled datasets is challenging or impractical. FSL leverages prior knowledge from related tasks or classes to make accurate predictions with minimal data.
In FSL, the model is typically trained on a variety of tasks, allowing it to learn a meta-representation of the data. When presented with a new task, the model can adapt quickly, utilizing the few available examples to make informed predictions. This adaptability is achieved through techniques such as meta-learning, transfer learning, and hybrid approaches.
A common framework in FSL is the “N-way K-shot” setup, where “N” represents the number of new classes, and “K” denotes the number of examples per class. For instance, in a 5-way 1-shot learning task, the model is presented with five new classes, each containing a single example, and must classify new instances into one of these classes.
« Back to Glossary Index