多输出分类综述

A Survey on Multi-output Classification

多输出分类(multi-output classification)旨在同时预测一个输入样例的多个离散输出值,其中多个输出可由多种离散变量(如二元变量、名义变量和有序变量)表征。近年来,该学习范式在机器学习领域已广受关注。尽管如此,现有研究工作主要针对输出结构简单的多输出分类任务(多类分类、多标签分类)展开研究,较少考虑现实应用中往往涉及复杂多样的输出空间结构,从而难以满足复杂多输出分类任务的需求。为了能够全面地应对各种变量表征输出的多种多输出分类任务,本文系统地阐述了不同多输出分类任务的定义,总结分析了各种多输出分类任务的输出空间结构特点以及问题建模方法,最后对多输出分类任务的发展方向进行了探讨。

Multi-output classification aims to predict multiple discrete outputs for an input,where the output values are characterized by diverse discrete variable types,such as binary,nominal and ordinal.In recent years,this learning paradigm has received extensive attentions in the field of machine learning.Nevertheless,prevailing researches mainly focus on multi-output classification tasks(multi-class classification,multi-label classification)with simple output structures,and few of them consider that diverse output space structures are often involved in the real-world applications,which makes it difficult to meet the needs of complex multi-output classification tasks.In order to deal with various multi-output classification tasks much better,this paper systematically expounds the definition of different multi-output classification tasks,summarizes both the characteristics of different output space structures and the modeling methods for various multi-output classification tasks.Finally,the development directions of multi-output classification tasks are discussed.