1: \begin{abstract}

2: We investigate the stochastic convex feasibility problem, the problem of finding $ x $ satisfying a system of constraints $ f_{\omega}(x) \leq 0 $ ($ \omega \in \Omega $), where the functions $ f_{\omega} $ are convex, and where $ \Omega $ is the sample space in a probability triple $ (\Omega, {\mathcal F}, P) $. Our aim is to solve the problem approximately, in the sense that given values $ \epsilon > 0 $ and $ 0 < \Gamma < 1 $, a point $ x $ is found for which the probability measure $ P $ satisfies $ P \{ \omega \in \Omega \mid f_{\omega}(x) \leq   \epsilon \} \geq 1 -  \Gamma  $. Our primary vehicle for obtaining such a point $ x $ is a subgradient method taking ``Polyak steps'', for which the only input is the size of minibatches to be employed in the effort. We prove stochastic iteration bounds for the general setting, and improved bounds when a notion of ``H\"{o}lderian growth'' is satisfied, showing, for example, linear convergence when the growth is linear (as happens, say, for a finite system of linear inequalities). We also derive deterministic iteration bounds for knowing ``with high confidence'' that an iterate $ x_k $ is in hand which fulfills the desired inequality, $ P\{ \omega \in \Omega \mid f_{\omega}(x_k) \leq   \epsilon \}  \geq 1 -  \Gamma  $. 

3: \end{abstract}