Few-shot learning

Home Glossary Item Few-shot learning
« Back to Glossary Index

Few-shot learning is a concept in machine learning designed to mimic human cognitive abilities by learning new concepts and making accurate predictions from a small number of examples – hence the term “few-shot”. Normal machine learning models usually require large amounts of data for training before they can make accurate predictions. Few-shot learning models aim to make accurate predictions or adapt to new classes not seen during training with just a handful of examples.


The technique is particularly important in instances where copious amounts of training data are not readily available. It also aims to make machine learning models more efficient by reducing the need for extensive data. This is achieved using methods like meta-learning or “learning to learn”, where models learn the underlying trend of the task from several related instances and then apply the learnt knowledge to new instances.

Few-shot learning comes from the need to build more efficient learning systems that can learn rapidly from a small number of examples, similarly to human cognition. It’s a step away from traditional extensive data-reliant systems towards more intelligent recognition systems, making it a significant concept in the advancement of artificial intelligence.

« Back to Glossary Index