Stochastic Approximation
Table of Contents
Overview
A simple goal: Find the solution to
![\begin{equation*}
\bar{f}(\theta^*) := \mathbb{E} \big[ f(\theta, W) \big] |_{\theta = \theta^*} = 0
\end{equation*}](../../assets/latex/stochastic_approximation_26a9cc9d48c208ba251f2edb71fb74e68fa8cf51.png)
Basic algorithm

The stepsize satisfies:
- Ensure we can reach anywhere:
- To attenuate noise:
Usually will take , which satisfies both of these.
Which we can write as:
![\begin{equation*}
\theta(n + 1) = \theta(n) + \alpha_n \big[ \bar{f} \big( \theta(n) \big) + \Delta (n + 1) \big]
\end{equation*}](../../assets/latex/stochastic_approximation_69ed37bae6790736cc93ee5c0eea575d61b385f9.png)
Which can be interpreted as a noisy Euler approximation to the ODE:
