机电工程技术2025，Vol.54Issue(6)：160-168,9.DOI:10.3969/j.issn.1009-9492.2024.00111
小样本学习研究综述
Review of Few-shot Learning Research
李坤 1陈剑钧 1李国胜 1姜晓道1
作者信息
- 1. 华东光电集成器件研究所,安徽 蚌埠 233042
- 折叠
摘要
Abstract
Deep neural networks require Numerous parameters are required by deep neural networks,necessitating a large amount of labeled data for model training.However,for many rare categories,a large number of labeled samples cannot be collected,which severely limits the scalability of deep learning methods.Inspired by remarkable human ability of few-shot learning,research in the field of few-shot learning has proliferated in recent years,yielding substantial progress.Methods in few-shot learning are systematically reviewed and summarized.First,the problem definition of few-shot learning is provided,the meta-learning framework for addressing few-shot problems is elaborated,and the specific processes of meta-training and meta-testing are explained.Then,a comprehensive overview of few-shot learning algorithms is provided,which are currently divided into three categories:optimization-based,memory-based,and metric-based methods,with the latter being the most widely applied and effective.A brief explanation of the first two types of methods was provided,followed by a focuse on the introduction of the metric-based methods.Finally,some possible future research directions in the field of few-shot learning were presented.关键词
小样本学习/元学习/度量学习Key words
few-shot learning/meta-learning/metric learning分类
信息技术与安全科学引用本文复制引用
李坤,陈剑钧,李国胜,姜晓道..小样本学习研究综述[J].机电工程技术,2025,54(6):160-168,9.
