SUBGRADIENT MINIMIZATION METHOD WITH DESCENT VECTORS CORRECTION BY MEANS OF TRAINING RELATIONS PAIRS
Autor: | V. N. Krutikov, Ya. N. Vershinin |
---|---|
Jazyk: | English<br />Russian |
Rok vydání: | 2014 |
Předmět: | |
Zdroj: | Вестник Кемеровского государственного университета, Vol 0, Iss 1-1, Pp 46-54 (2014) |
Druh dokumentu: | article |
ISSN: | 2078-8975 2078-8983 |
Popis: | The paper introduces a conjugate subgradient method whose descent is corrected by a pair of current training relations. The convergence of the method is proved on strictly convex functions. According to the numerical experiment, the method is effective at non-smooth high-dimensional minimization problems. By memory cost, theproposed method is similar to the conjugate gradient method, and at smooth high-dimensional singular functions its convergence rate is not inferior to tha of the conjugate gradient method. |
Databáze: | Directory of Open Access Journals |
Externí odkaz: |