GFlowNets for combinatorial optimization through Markov decision processes

This method utilizes GFlowNets to sample from composite unnormalized densities, addressing NP-hard combinatorial optimization problems. By designing specific MDPs and using conditional GFlowNets, the approach enables efficient sampling and diverse candidate generation in highly constrained solution spaces.