现代信息科技2025，Vol.9Issue(14)：37-41,5.DOI:10.19850/j.cnki.2096-4706.2025.14.008
基于双重约束的非线性跨领域特征学习方法
Nonlinear Cross-domain Feature Learning Method Based on Double Constraints
摘要
Abstract
In recent years,Deep Learning models have been widely used in cross-domain classification learning.Marginalized Denoising Autoencoder(mDA)is widely used because of its linear structure to obtain the optimal solution and fast training speed.However,mDA does not describe the nonlinear relationship of feature space in feature representation learning,and does not fully consider the distribution difference between domains when constructing a new feature space.Therefore,this paper proposes a Cross-Domain Feature Representation Learning Method Based on Nonlinear Dual-Constraints(FNDC).This method introduces a kernel function in the reconstruction error to capture the nonlinear relationship of the feature space.At the same time,the Maximum Mean Difference(MMD)and Manifold Regularization(MR)are used as dual constraints to reduce the cross-domain distribution differences and maintain the relative position relationship of the space after the feature mapping.The experimental results show that the proposed method performs better than the existing baseline algorithms in cross-domain text classification tasks.关键词
跨领域/最大均值差异/流形正则化/非线性Key words
cross-domain/MMD/MR/nonlinear分类
信息技术与安全科学引用本文复制引用
丁晗..基于双重约束的非线性跨领域特征学习方法[J].现代信息科技,2025,9(14):37-41,5.
