state abstractions in reinforcement learning

Exploiting graph structure in the MDP for learning state abstractions