ALGORITM CU MODIFICAREA ALEATORIE A COMPONENTELOR GRADIENTULUI PENTRU UN MODEL CONVEX DE OPTIMIZARE
Pavel BALAN Catedra Informare şi Optimizare Discretă
Autori
USM ADMIN
Studia Universitas Moldaviae
Rezumat
A stochastic algorithm is proposed and analyzed, that is a probabilistic generalization of gradient method, for
solving convex models. A random change of „old” partial derivatives with „new” ones is performed from one iteration
to another. Convergence aspects of this scheme are analyzed for the case when the step is adjusted programmatically.
Certain conditions are indicated, that, being respected, ensure the convergence of this scheme to the optimal solution
with probability 1.
