Few-shot Learning using Data Augmentation: A Literature Review

Abstract

This paper examines few-shot learning, a machine-learning approach that allows models to generalize well with scarce labeled data. It overviews few-shot learning concepts and reviews data augmentation techniques tailored for this scenario. It evaluates these image and object recognition techniques and discusses their benefits and limitations. An experimental setup is described, with selected datasets and metrics to assess the augmentation strategies. Baseline models are analyzed, and the comparative results are presented, identifying key trends and areas for further research in few-shot learning.