关于分类问题中回归模型的教学探讨  

A Teaching Discussion on Regression Models in Classification Problems

作　　者：贾彬彬 张爱华[1] 张敏灵[2] JIA Binbin;ZHANG Aihua;ZHANG Minling(College of Electrical and Information Engineering,Lanzhou University of Technology,Lanzhou 730050,China;School of Computer Science and Engineering,Southeast University,Nanjing 210096,China)

机构地区：[1]兰州理工大学电气工程与信息工程学院,兰州730050 [2]东南大学计算机科学与工程学院,南京210096

出　　处：《电气电子教学学报》2024年第6期128-131,共4页Journal of Electrical and Electronic Education

基　　金：兰州理工大学研究生课程思政建设项目(KCSZ-YB-202242013);江苏省高等教育教改重点课题(2021JSJG104)。

摘　　要：回归与分类是完全不同的学习任务,前者旨在拟合样本的实值标记,而后者旨在将不同类别的样本分开。然而,通过拟合样本的二值标记来求解分类问题亦常可见。首先,对回归和分类的线性模型进行了分析;然后,通过分析基于回归的分类模型,指出基于回归的分类方法实际上是使用平方损失作为理想0/1损失的替代损失函数;最后,讨论了教学安排、课程思政以及可进一步研究的问题。Regression and classification are two different learning tasks.The former aims to fit the real-valued labels of samples,while the latter aims to separate samples of different classes.However,it is also common to solve classification problems by fitting binary labels of samples.Firstly,the linear models of regression and classification tasks are analyzed.Then,by analyzing the regression-based classification models,it is pointed out that the regression-based classification method actually uses the square loss as a surrogate loss function for the ideal 0/1 loss.Finally,the teaching arrangement,the curriculum ideology and politic,and the problems that can be further studied are discussed.

关 键 词：分类问题 线性回归 替代损失 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]