Tag
The article presents BFN-RL, a unified generative modeling framework for offline reinforcement learning based on Bayesian Flow Networks, capable of generating effective trajectories across discrete and continuous state spaces.