Shen et al.
Stochastic attention routing enables efficient RL alignment for diffusion models
Routing Policy Optimization (RPO) samples attention routes from the generator's own attention mass at selected denoising levels, using a two-rollout advantage estimator to update the dense generator without modifying the original sampler.
Chinese Tech
Qiuhong Shen · Yiding Yang · Yuxin Zhang · Xinwei Huang · Minxuan Lin · Bo Liu · +5 more
National University of Singapore · ByteDance
Research Digest··3 min read
The authors propose Routing Policy Optimization (RPO), a method that introduces stochastic exploration into diffusion and flow generators by randomly sampling attention routes during training while preserving the original deterministic sampler at inference.
Why this paper
From ByteDance and National University of Singapore
In one line
Routing Policy Optimization uses temporary stochastic attention routes to improve reward-aligned visual generation with less training compute while preserving dense deterministic inference.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§