AN ALTERNATIVE PROOF FOR CONVERGENCE OF STOCHASTIC-APPROXIMATION ALGORITHMS

Authors
KULKARNI SR HORN CS
Citation
Sr. Kulkarni et Cs. Horn, AN ALTERNATIVE PROOF FOR CONVERGENCE OF STOCHASTIC-APPROXIMATION ALGORITHMS, IEEE transactions on automatic control, 41(3), 1996, pp. 419-424
Citations number
14
Categorie Soggetti
Controlo Theory & Cybernetics","Robotics & Automatic Control","Engineering, Eletrical & Electronic
Journal title
IEEE transactions on automatic control → ACNP
ISSN journal
00189286
Volume
41
Issue
3
Year of publication
1996
Pages
419 - 424
Database
ISI
SICI code
0018-9286(1996)41:3<419:AAPFCO>2.0.ZU;2-X
Abstract
An alternative proof for convergence of stochastic approximation algor ithms is provided. The proof is completely deterministic, very element ary (involving only basic notions of convergence), and direct in that it remains in a discrete setting. An alternative form of the Kushner-C lark condition is introduced and utilized and the results are the firs t to prove necessity for general gain sequences in a Hilbert space set ting.