Michael Thon

• Input-output OOMs (IO-OOMs) are an extension of the basic OOM theory to deal with controlled stochastic processes. They were first presented in a comprehensive OOM tutorial text, where an early version of the OOM learning algorithm was partially transferred. Since then, OOM research has focussed on basic OOMs, in particular on developing statistically efficient learning algorithms, and research on IO-OOMs has been dormant. In this report it is shown how the main theorems for OOMs can be transferred to the case of IO-OOMs and how one of the current OOM learning algorithms can be adapted to learning IO-OOMs, yielding the first complete IO-OOM learning algorithm.