摘要
为了提高二线性迭代最小二乘(BALS)算法拟合平行因子(PARAFAC)模型的速度,提出了一种新的PARAFAC模型拟合算法.该算法利用新迭代与旧迭代之间的增量值,来预测下一次迭代的初始值,对BALS中的每次迭代,为2个加载矩阵设置相应的松弛因子,并通过联合优化的方法求得最优松弛因子对,从而加速BALS的收敛速度.理论分析和仿真结果表明,与已有的BALS算法相比,所提算法在不牺牲性能的条件下,有效地提高了PARAFAC模型的拟合速度.
To speed up the convergence of the bilinear alternating least squares （BALS） algorithm of fit- ting the parallel factor （PARAFAC） model, a new algorithm of fitting the PARAFAC model was pro- posed. In each iteration, the proposed algorithm sets up their own relaxation factors for two loading matri- ces which are required to be estimated, and gets the optimal couple of two relaxation factors by the joint optimization. Analysis and simulation show that the proposed algorithm improves the speed of fitting the PARAFAC model without performance deterioration compared with the existing BALS algorithm.
出处
《北京邮电大学学报》
EI
CAS
CSCD
北大核心
2014年第4期29-33,共5页
Journal of Beijing University of Posts and Telecommunications
基金
国家高技术研究发展计划项目(2014AA01A701)
国家自然科学基金项目(60872149)
关键词
二线性迭代最小二乘
平行因子
迭代
松弛因子
收敛
bilinear alternating least squares
parallel factor
iteration
relaxation factor
convergence
