A stochastic optimal control problem can be interpreted as a free energy minimization problem [0], but this is a more general result than what is looked at in t

A stochastic optimal control problem can be interpreted as a free energy minimization problem [0], but this is a more general result than what is looked at in the paper OP linked. I'm not sure the "reverse SDE" is technically doing this sort of minimization since it seems like it's just trying to reconstruct an unknown forward process. It's possible there's some sort of minimization going on here over some slack variable though. [0] https://ieeexplore.ieee.org/abstract/document/6426381?casa_token=LFlej7BJyesAAAAA:H48riM4Ivo3eKX82_tMxxBWYGkofMR3hVXZBB87cGEbll0ecyAxQye1jIYSOyL82MoXX6BRcnA https://ieeexplore.ieee.org/abstract/document/6426381?casa_t...