[1]李平,仝涛.面向分类任务的鲁棒高斯过程隐变量模型[J].计算机技术与发展,2025,(02):86-92.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0316]
　LI Ping,TONG Tao.Robust Gaussian Process Latent Variable Model for Classification Task[J].,2025,(02):86-92.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0316]
点击复制

面向分类任务的鲁棒高斯过程隐变量模型()

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2025年02期
页码:
86-92
栏目:
人工智能
出版日期:
2025-02-10

文章信息/Info

Title:
Robust Gaussian Process Latent Variable Model for Classification Task
文章编号:
1673-629X(2025)02-0086-07
作者:
李平; 仝涛
南京邮电大学 计算机学院,江苏 南京 210023
Author(s):
LI Ping; TONG Tao
School of Computer Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,China
关键词:
高斯过程; 隐变量模型; 监督学习; 离群点噪声; 鲁棒学习
Keywords:
Gaussian process; latent variable model; supervised learning; outlier noise; robust learning
分类号:
TP311
DOI:
10.20165/j.cnki.ISSN1673-629X.2024.0316
摘要:
高斯过程隐变量模型(GPLVM)作为一种高效的贝叶斯无参降维模型,目前已广泛应用于多种应用中。 然而传统的 GPLVM 模型为完全无监督降维模型,导致其无法有效嵌入监督信息,从而极大地限制了其在监督学习任务中的性能。同时 GPLVM 通常假设数据噪声服从高斯分布,因此无法建模离群点噪声,从而严重影响了模型的鲁棒性。 针对上述问题,该文提出一种面向分类任务的鲁棒高斯过程隐变量模型。 该模型通过引入基于原型的分类器并基于此构建隐变量先验分布,实现了有监督的数据降维和分类。 同时,通过利用均值来建模数据的离群点噪声,进而实现了模型的鲁棒学习。在多个数据集上的实验结果表明,该模型在嵌入监督信息同时,可对离群点噪声进行有效建模,从而获得更优的分类准确率。
Abstract:
Gaussian process latent variable model (GPLVM) as an effective Bayesian non-parametric model,has been widely used in many applications. However,since the conventional GPLVM is a fully unsupervised dimension reduction model,it cannot effectively embed the supervised information,limiting the application scope in the supervised learning tasks. Moreover,GPLVM assumes that the noise of data follows a Gaussian distribution,thus it cannot model the outlier noise,which reduces the robustness of the model. To address these problems,we propose a robust Gaussian process latent variable model for the classification task. This model utilizes the prototype based classifier and constructs a prior distribution of latent variables,implementing the supervised dimension reduction and classification.Furthermore,it uses the mean of Gaussian process to model the outlier noise and realize the robust learning of model. The experimental results on multiple dataset demonstrate the proposed model can effectively embed the supervised information and capture the outlier noise,obtaining more superior classification performance.

相似文献/References:

[1]张 峰.基于近似边界和聚类的昂贵多目标优化算法[J].计算机技术与发展,2022,32(S1):21.[doi:10. 3969 / j. issn. 1673-629X. 2022. S1. 005]
　ZHANG Feng.Expensive Multi-objective Optimization Algorithm Based on Approximate Boundary and Clustering[J].,2022,32(02):21.[doi:10. 3969 / j. issn. 1673-629X. 2022. S1. 005]
[2]王文豪,李秀芹.GPAformer模型在气温预测中的应用研究[J].计算机技术与发展,2025,(03):187.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0359]
　WANG Wen-hao,LI Xiu-qin.Research on Application of a GPAformer Model for Temperature Prediction[J].,2025,(02):187.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0359]

更新日期/Last Update: 2025-02-10