||Hi! I would like to share the idea with how to improve the calculation of the value function of dynamic programs. I am a mathematician and view dynamic programming from the mathematical point of view. If you have profound knowledge of dynamic programming you have no choice but to be familiar with value functions. In approximate dynamic programming (or adaptive dynamic programming because it's the same), we use statistical methods to estimate value functions such as regression of methods of instrumental variables. What I offer is to use Random Forests instead of regression. I do research in this field, i.e. to estimate value functions using random forests and hope to have results soon.
Vadym Omelchenko. Charles University in Prague. PhD student.