Can any dynamic programming problem be formulated as simulation optimization?
I was wondering that because I am researching the use ML in optimization problems (and not the other way around) and found out the ML applications to SO - basically we can train an ML model to reproduce the SO model, and then we can perform the optimization on the ML model. However, I am struggling to understand how ML can be used directly in DP problems - and thought this would maybe be a solution (DP applicable problem formulation (a problem that can be solved with DP -> SO (model the problem as SO) -> ML model and then use DP method to solve the ML model)
Does it make any sense? What would you be your recommendations on how to use ML in DP problems?