RPO: Routing Policy Optimization in Online Diffusion Reinforcement Learning

Routing Policy Optimization (RPO) has been introduced to enhance the training efficiency of online diffusion reinforcement learning. It incorporates random attention routing, achieving nearly 4x higher training efficiency compared to the Flow-GRPO method in tasks like text-to-image and text-to-video generation. This development is a significant advancement in the field of AI and machine learning.

AI 根据报道生成 · 1 小时前更新

还没有足够的连续观测数据，暂不绘制趋势。