摘要
为了提高高维数据维数约简的计算效率,基于局部邻域相关的权重与稀疏矩阵,提出了1种改进的局部线性嵌入算法。对于高维数据维数约简的信息量估计,采用了相关维数估计方法来计算一致流形信息量的上界。采用Swiss、Broken swiss、Helix、Twinpeaks和Intersect 5种经典数据集进行实验评估。实验结果显示,与局部线性嵌入算法相比,针对5种经典数据集,该文算法速度分别提高了27.60%、27.51%、27.18%、28.31%和45.28%。
An improved locally linear embedding(LLE) algorithm based on local neighborhood-dependent weights and sparse matrices is proposed to improve the computation efficiency of dimensionality reduction for high-dimensional data.The correlation dimension estimation method is used to estimate the intrinsic information of the dimensionality reduction in high-dimensional data and the upper bound of the uniform manifold. Five classical datasets,including Swiss,Broken swiss,Helix,Twinpeaks and Intersect,are used to assess the algorithm. The results show that,compared with that of local linear embedding algorithm,the calculation speed of this algorithm on the five datasets is improved by 27.60%,27.51%,27.18%,28.31% and 45.28% respectively.
出处
《南京理工大学学报》
EI
CAS
CSCD
北大核心
2017年第6期748-752,共5页
Journal of Nanjing University of Science and Technology
基金
江苏省高等学校自然科学研究项目(17KJB470002)
关键词
自适应邻域选择
局部线性嵌入
稀疏矩阵
数据降维
流形算法
adaptive neighborhood selection
locally linear embedding
sparse matrices
dimension reduction
manifold algorithm
