当前位置:主页 > 医学论文 > 护理论文 >

用于微阵列数据分类的子空间融合演化超网络

发布时间:2018-04-25 15:11

  本文选题:模式识别 + 微阵列数据分类 ; 参考:《电子学报》2016年10期


【摘要】:针对传统模式识别方法在学习具有小样本特性的DNA微阵列数据时存在的过拟合问题,本文提出了一种子空间融合演化超网络模型.该模型通过子空间划分、超边全覆盖和子空间融合三种方法降低模型对初始化的依赖,减少了对数据空间的拟合误差,提高了演化超网络的泛化能力.对四个DNA微阵列数据集的实验结果表明,子空间融合演化超网络的识别率和在小样本训练集下的泛化能力均优于参与对比的其他传统模式识别方法.
[Abstract]:Aiming at the over-fitting problem of traditional pattern recognition methods in learning DNA microarray data with small sample characteristics, a subspace fusion evolutionary supernetwork model is proposed in this paper. The model reduces the dependence on initialization, reduces the fitting error of data space and improves the generalization ability of evolutionary supernetwork by subspace partitioning, hyper-edge full coverage and subspace fusion. The experimental results of four DNA microarray datasets show that the recognition rate of the subspace fusion evolution supernetwork and the generalization ability under the small sample training set are superior to those of other traditional pattern recognition methods involved in the comparison.
【作者单位】: 重庆邮电大学计算智能重庆市重点实验室;
【基金】:国家自然科学基金(No.61203308,No.61403054) 重庆教委科学技术研究项目(自然科学类)(No.KJ1400436) 重庆市基础与前沿研究计划项目(No.cstc2014jcyj A40001)
【分类号】:R440;O157.5


本文编号:1801854

资料下载
论文发表

本文链接:https://www.wllwen.com/huliyixuelunwen/1801854.html


Copyright(c)文论论文网All Rights Reserved | 网站地图 |

版权申明:资料由用户61318***提供,本站仅收录摘要或目录,作者需要删除请E-mail邮箱bigeng88@qq.com