计算机工程

• 人工智能及识别技术 • 上一篇    下一篇

基于非线性嵌入的自联想神经网络

吴昊东   

  1. (复旦大学 计算机科学技术学院, 上海 201203)
  • 收稿日期:2016-03-01 出版日期:2017-07-15 发布日期:2017-07-15
  • 作者简介:吴昊东(1991—),男,硕士研究生,主研方向为机器学习、神经网络。
  • 基金项目:
    教育部高等学校博士学科点专项科研基金(20120071110035)。

Auto-associative Neural Network Based on Nonlinear Embedding

WU Haodong   

  1. (School of Computer Science,Fudan University,Shanghai 201203,China)
  • Received:2016-03-01 Online:2017-07-15 Published:2017-07-15

摘要: 传统分类器常依赖于低维度子空间的特征进行分类,但仅在单个子空间下进行分类会因为不同类别的重叠而效果不佳。为此,提出一种基于流形学习的神经网络分类方法,利用非线性嵌入方法获得数据每个类的子空间,再使用非线性嵌入判别准则优化各个径向基函数自联想神经网络的参数。实验结果表明,该方法能有效解决类别重叠问题,分类准确率和鲁棒性高于传统分类方法。

关键词: 非线性嵌入, 自联想神经网络, 径向基函数, 深度学习, 模式识别

Abstract: Most classifiers prefer to classify high-dimensional data based on the characteristics of low-dimensional subspace.However,its performance suffers from overlapping among different classes in the single subspace.To address this issue,a classification method using manifold learning based neural network is proposed.This method uses nonlinear embedding to attain the subspace of each class,and nonlinear embedding criterion to optimize the parameters of each basis function auto-associative neural network.Experimental results show that this method is more robust and has higher accuracy than traditional methods.

Key words: nonlinear embedding, auto-associative neural network, radial basis function, deep learning, pattern recognition

中图分类号: