Summary
2023
Session Number:D2L-2
Session:
Number:D2L-23
A Novel HALS-Based Iterative Algorithm for Randomized Nonnegative Matrix Factorization
Masuda Takao, Yamada Kento, Migita Tsuyoshi, Takahashi Norikazu,
pp.704-707
Publication Date:2023-09-21
Online ISSN:2188-5079
Summary:
As an approach to efficiently perform large-scale Nonnegative Matrix Factorization (NMF), randomized NMF was recently proposed. This approach reduces the dimension of a given nonnegative matrix by multiplying it by a random matrix, and then performs matrix factorization. However, algorithms for standard NMF cannot be directly used in randomized NMF because linear inequality constraints have to be considered instead of nonnegativity ones. In this paper, we reformulate the optimization problem for randomized NMF from a different perspective and propose a novel iterative algorithm which is a combination of the hierarchical alternating least squares algorithm and projection onto the feasible region. We also prove that the proposed algorithm has global convergence property.