Read the original at HF Daily Papers
Routing Policy Optimization introduces stochastic exploration into diffusion models by sampling attention routes during denoising, enabling reinforcement learning alignment without altering the original sampler.
Carried by: HF Daily Papers. First seen: .