【摘要】 给出一种广义的Kullback-Leibler代价函数,基于调比梯度下降法得到新的非负矩阵分解算法。新算法的优点是能够对稀疏非负矩阵进行分解,但是新算法的收敛性没有得到改善。进一步对新算法进行改进,数值实验表明改进后算法的收敛性得到明显改善。更多还原

【Abstract】 We presented a generalized Kullback-Leibler cost function,and derived a new nonnegative matrix factorization algorithm based on scaled gradient desent method.The merit of the new algorithm lies in that it could decompose sparse nonnegative matrix,and the defect is that the convergence has not been improved.We improved further the new algorithm,and the numercial experiment indicate the convergence of the improved algorithm has been improved well.更多还原