Research On NMF Algorithms

In this thesis,we focus on the algorithm of NMF. We propose the subspace conjugate gradient algorithm and the mixed conjugate gradient algorithm.Firstly, We introduce the background knowledge,research status applications and make a brief introduction of classical NMF algorithm, and gives an analysis for all the proposed algorithms.Secondly, Based on the alternating nonnegative least squares (ANLS) framework and the accurate active set conjugate gradient algorithm for bound constrained optimization, the author proposes a subspace conjugate gradient algorithm to solve the NMF problem. The author show numerical experiments.With comparison to PG algorithm, AS and BPP algorithm, our new algorithm shows advantages in time and alteration.Finally, for the search direction of step3of section of2.3, the author adopts the search direction of mixed conjugate gradient algorithm of unconstrained optimization of ming yuan Cao. Numerical experiments show that new algorithm have an advantage in time and iterations.