The convergence of stochastic gradient algorithms applied to learning in neural networks. (English. Russian original) Zbl 1060.60501

Stanković, S.; Tadić, V.

Autom. Remote Control 59, No. 7, Part 2, 1002-1015 (1998); translation from Avtom. Telemekh. 1998, No. 7, 118-134 (1998).

Summary: The results of convergence of stochastic gradient algorithms obtained under general assumptions are applied to study supervised learning algorithms in feedforward neural networks. Sufficient conditions for the local convergence of parameter sequences and almost-sure global and mean-square convergences of the criterion sequence are derived.

MSC:

60F15	Strong limit theorems
60F25	\(L^p\)-limit theorems
93E25	Computational methods in stochastic control (MSC2010)
68T05	Learning and adaptive systems in artificial intelligence