摘要
Few‐shot learning is an emerging research field in recent years, aiming at solving machine learning tasks with limited samples.Since a large number of samples cannot be obtained for many scenarios in the real world, such as lung cancer in the medical field, the method of few‐shot learning is popular in many fields. However, due to the limitation of sample size, the problem of model overfitting is often caused. This paper first describes the definition of few‐shot learning, systematically sorts out the current work related to few‐shot learning, and specifically introduces the research progress of three types of few‐shot learning models based on data enhancement, transfer learning and meta‐learning. Then the siamese network, the model‐agnostic meta‐learning and the prototype network are studied in three small samples The classical method is described in detail. Finally, the future development direction of few‐shot learning is prospected.