投影Barzilai-Borwein法求解非負矩陣分解

Projected Barzilai-Borwein Methods for Non-negative Matrix Factorization

蔡鎮澤(Chen-Tse Tsai)

指導教授 ： 林智仁

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2011年)

https://doi.org/10.6342/NTU.2011.01128

摘要

無資料

關鍵字

非負矩陣分解 ； Barzilai-Borwein法 ； 梯度下降法

並列摘要

Non-negative matrix factorization (NMF) is a useful dimension reduction tech- nique. Currently, the most effective way to minimize NMF optimization problems is by alternatively solving non-negative least square sub-problems. Some recent stud- ies have shown that projected Barzilai-Borwein methods are very efficient for solving each sub-problem. In this thesis, we study variants of the projected Barzilai-Borwein methods and discuss some useful implementation techniques. We provide an efficient implementation to succeed our popular NMF code via a projected gradient method.

並列關鍵字

Non-negative matrix factorization ； Barzilai-Borwein method ； projected gradient method.

參考文獻

J. Barzilai and J. M. Borwein. Two-point step size gradient methods. IMA Journal of Numerical Analysis, 8:141–148, 1988.

D. P. Bertsekas. On the Goldstein-Levitin-Polyak gradient projection method. IEEE Transactions on Automatic Control, 21:174–184, 1976.

E. G. Birgin, J. M. Martínez, and M. Raydan. Nonmonotone spectral projected gradient methods on convex sets. SIAM Journal on Optimization, 10:1196–1211, 2000.

S. Bonettini. Inexact block coordinate descent methods with application to non- negative matrix factorization. IMA Journal of Numerical Analysis, 2011. To appear.

P. H. Calamai and J. J. Moré. Projected gradient methods for linearly constrained problems. Mathematical Programming, 39:93–116, 1987.