少样本学习  

Few-shot learning

作　　者：张艺超 侯艳杰 陈君华[4] 唐轶[4] ZHANG Yi-chao;Hou Yan-jie;CHEN Jun-hua;Tan Yi(Key Laboratory of Spectral Imaging Technology,Xi'an Institute of Optics and Precision Mechanics,Chinese Academy of Sciences,Xi'an 710119,China;University of Chinese Academy of Sciences,Beijing 100049,China;Taiyuan Satellite Launch Center,Taiyuan 030027,China;Key Laboratory of IOT Application Technology of Universities in Yunnan Province,Yunnan Minzu University,Kunming 650500,China)

机构地区：[1]中国科学院西安光学精密机械研究所光谱成像技术重点实验室,陕西西安710119 [2]中国科学院大学,北京100049 [3]太原卫星发射中心,山西太原030027 [4]云南民族大学云南省高校物联网应用技术重点实验室,云南昆明650500

出　　处：《云南民族大学学报（自然科学版）》2020年第6期582-591,共10页Journal of Yunnan Minzu University:Natural Sciences Edition

基　　金：国家自然科学基金(61866040)。

摘　　要：人工智能在很多领域都得到了迅速发展,但现有方法需要在大量的数据中学习先验知识.为了进一步缩小人工智能与人类差距,使其可以从少量的监督信息中学习,获得在新任务上的泛化能力,出现了少样本学习方法.少样本学习的目的是利用少量的有标签样本学习一个分类器,对未知的类进行识别.本文对少样本学习方法的概念和应用场景进行了概述,讨论了诸如半监督学习、数据不平衡学习、迁移学习和元学习之类的相关学习问题与少样本学习间的关联.本文对主流少样本学习方法进行了系统的介绍,通过全面比较将其归类为不同类别.最后,展示了一些主流少样本学习方法在分类任务上的实验结果并加以分析.Artificial intelligence has developed rapidly in many areas, but existing methods require large amounts of data. In order to further reduce the gap between artificial intelligence and humans, so that they can learn from limited supervised information to gain generalization ability on new tasks, there is a kind of learning called few-shot learning. The purpose of few-shot learning is to learn a classifier that can identify unseen classes with a small number of labeled samples. This paper gives an overview of the concepts and application scenarios of the few-shot learning method, the relations between few-shot learning and related learning problems are discussed such as semi-supervised learning, data-imbalanced learning, transfer learning, and meta-learning. This paper systematically introduces the mainstream few-shot learning methods, classifies them into different categories through comprehensive comparison. Experimental results are also presented and analysed.

关 键 词：少样本学习 迁移学习 元学习 域漂移 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]