Stochastic approximation: Difference between revisions

Content deleted Content added
Line 12:
* <math>M(x)</math> is nondecreasing,
* <math>M'(\theta)</math> exists and is positive, and
* <math>a_n</math> satisfies the following requirements <math>\quadqquad \sum^{\infty}_{i=0}a_i = \infty \quad </math> and <math>\quad \sum^{\infty}_{i=0}a^2_i < \infty \quad </math>
 
The last condition is fulfilled for example by taking <math>a_n=1/n</math> for example; other series are possible but in order to average out the noise in <math>N(x)</math>, <math>a_n</math> must converge slowly.
 
==Kiefer-Wolfowitz algorithm==