An Experimental Study of Balance in Matrix Factorization
| Autor: | Peter J. Ramadge, Jennifer Hsia |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
Linear programming
010501 environmental sciences 01 natural sciences Critical point (mathematics) Matrix decomposition 03 medical and health sciences 0302 clinical medicine Factorization Saddle point Convergence (routing) Applied mathematics Gradient descent 030217 neurology & neurosurgery Subspace topology 0105 earth and related environmental sciences Mathematics |
| Zdroj: | CISS |
| DOI: | 10.1109/ciss50987.2021.9400232 |
| Popis: | We experimentally examine how gradient descent navigates the landscape of matrix factorization to obtain a global minimum. First, we review the critical points of matrix factorization and introduce a balanced factorization. By focusing on the balanced critical point at the origin and a subspace of unbalanced critical points, we study the effect of balance on gradient descent, including an initially unbalanced factorization and adding a balance-regularizer to the objective in the MF problem. Simulations demonstrate that maintaining a balanced factorization enables faster escape from saddle points and overall faster convergence to a global minimum. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|