This paper provides a comprehensive survey of Few-Shot Learning (FSL), a machine learning paradigm that aims to enable models to learn from a small number of examples with supervised information. The authors define FSL formally and distinguish it from other machine learning problems. They identify the core issue in FSL as the unreliability of the empirical risk minimizer due to the limited number of training samples. To address this issue, they categorize FSL methods into three perspectives: data, model, and algorithm. Data-augmentation methods enhance the training set by adding more samples, while model-augmentation methods reduce the hypothesis space to make learning feasible. Algorithm-augmentation methods alter the search strategy to find the best hypothesis. The paper reviews and discusses the pros and cons of each category, proposes future directions, and provides insights for future research.This paper provides a comprehensive survey of Few-Shot Learning (FSL), a machine learning paradigm that aims to enable models to learn from a small number of examples with supervised information. The authors define FSL formally and distinguish it from other machine learning problems. They identify the core issue in FSL as the unreliability of the empirical risk minimizer due to the limited number of training samples. To address this issue, they categorize FSL methods into three perspectives: data, model, and algorithm. Data-augmentation methods enhance the training set by adding more samples, while model-augmentation methods reduce the hypothesis space to make learning feasible. Algorithm-augmentation methods alter the search strategy to find the best hypothesis. The paper reviews and discusses the pros and cons of each category, proposes future directions, and provides insights for future research.