北京印刷学院学报
北京印刷學院學報
북경인쇄학원학보
JOURNAL OF BEIJING INSTITUTE OF GRAPHIC COMMUNICATION
2015年
4期
53-58
,共6页
非负矩阵分解%稀疏约束%曼哈顿距离%欧几里得距离
非負矩陣分解%稀疏約束%曼哈頓距離%歐幾裏得距離
비부구진분해%희소약속%만합돈거리%구궤리득거리
non-negative matrix factorization%sparse constraints%Manhattan distance%Euclidean distance
非负矩阵分解是处理高维数据的一种常用方法,对带有稀疏约束的非负矩阵分解算法进行了研究,提出了一种在曼哈顿距离最接近的向量稀疏化算法,并与欧几里得距离最接近的向量稀疏化进行对比,提出的稀疏化算法具有较快的稀疏速度和较好的稀疏效果。实验结果表明,只在非负基矩阵 W 上加稀疏约束时,得到的非负基矩阵 W和非负系数矩阵 H 的乘积和原非负矩阵 V 最接近;基于稀疏的非负矩阵分解过程中,选取的迭代步长和迭代次数,会对实验分解结果产生较大影响。
非負矩陣分解是處理高維數據的一種常用方法,對帶有稀疏約束的非負矩陣分解算法進行瞭研究,提齣瞭一種在曼哈頓距離最接近的嚮量稀疏化算法,併與歐幾裏得距離最接近的嚮量稀疏化進行對比,提齣的稀疏化算法具有較快的稀疏速度和較好的稀疏效果。實驗結果錶明,隻在非負基矩陣 W 上加稀疏約束時,得到的非負基矩陣 W和非負繫數矩陣 H 的乘積和原非負矩陣 V 最接近;基于稀疏的非負矩陣分解過程中,選取的迭代步長和迭代次數,會對實驗分解結果產生較大影響。
비부구진분해시처리고유수거적일충상용방법,대대유희소약속적비부구진분해산법진행료연구,제출료일충재만합돈거리최접근적향량희소화산법,병여구궤리득거리최접근적향량희소화진행대비,제출적희소화산법구유교쾌적희소속도화교호적희소효과。실험결과표명,지재비부기구진 W 상가희소약속시,득도적비부기구진 W화비부계수구진 H 적승적화원비부구진 V 최접근;기우희소적비부구진분해과정중,선취적질대보장화질대차수,회대실험분해결과산생교대영향。
Non-negative matrix factorization ( NMF ) is commonly used method for processing high-dimensional data. In this paper,we have a research on NMF with sparse constraints,and propose a sparse vector algorithm have the closest Manhattan distance,compared with the sparse vector algorithm have the closest Euclidean distance,the algorithm has faster speed and better sparse results. Experimental results show that only the non-negative basis matrix W with sparse constraints resulting non-negative basis matrix W multiplied by the non-negative coefficient matrix H is closest to the original non-negative matrix V. In the process of NMF with sparse, iterative step length and the number of iterations selected,will have a greater impact experiments decomposition results.