Title :
Adjacent errors of stochastic gradient algorithms under general error criteria
Author :
Anant, Venkat ; Priemer, Roland
Author_Institution :
Dept. of Electr. Eng. & Comput. Sci., Illinois Univ., Chicago, IL, USA
Abstract :
This paper gives a method to distinguish between the transient and steady states of stochastic gradient algorithms under general error criteria using inequalities that exist among a set of output errors called adjacent errors. Based on these inequalities, a variable step size LMF algorithm is proposed which improves convergence and gives lower misadjustment
Keywords :
encoding; filtering theory; identification; prediction theory; transient analysis; adjacent errors; convergence; error criteria; misadjustment; output errors; steady states; stochastic gradient algorithms; transient states; variable step size LMF algorithm; Computer errors; Convergence; Equations; Estimation error; Least squares approximation; Steady-state; Stochastic processes; Stochastic resonance; Stochastic systems; White noise;
Conference_Titel :
Circuits and Systems, 1997. Proceedings of the 40th Midwest Symposium on
Conference_Location :
Sacramento, CA
Print_ISBN :
0-7803-3694-1
DOI :
10.1109/MWSCAS.1997.662190
