增量学习研究综述  

A Review of Incremental Learning

作　　者：姚越 周仁杰 张伟[1] YAO Yue;ZHOU Renjie;ZHANG Wei(School of Computer Science,Hangzhou Dianzi University,Hangzhou Zhejiang 310018)

机构地区：[1]杭州电子科技大学计算机学院,浙江杭州310018

出　　处：《软件》2023年第12期29-32,37,共5页Software

基　　金：国家重点研发计划项目(2022YFB3105401)。

摘　　要：近年来,机器学习越来越多地应用于各种生产和生活场景中。然而,在现实中,新的数据不断产生,学习到新知识的同时会忘记旧的知识,导致灾难性遗忘,而增量学习被用来解决该问题。本文介绍了增量学习的基本概念,以及三种主流方法的特点及代表性方法,并对当前增量学习遇到的问题进行总结,最后对未来的研究方向做出展望。In recent years,machine learning has been increasingly used in various production and life scenarios.However,in reality,new data is constantly being generated.Old knowledge will be forgotten while learning new knowledge,leading to catastrophic forgetting.Incremental learning is used to solve this problem.This article introduces the basic concepts of incremental learning,discusses the characteristics and representative methods of three mainstream approaches.It also summarizes the challenges faced by current incremental learning methods and provides a prospective outlook on future research directions.

关 键 词：机器学习 增量学习 在线学习 灾难性遗忘 正则化方法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]