Preprint
Aug 2026
Disagree to Accelerate: Closing the Loop on Diffusion Feature Forecasts
This work introduces RACER, a training-free closed-loop controller that continuously shrinks uncertain forecasts toward the last computed feature and derives a deterministic error bound for the shrinkage and empirically evaluates its validity and tightness across acceleration regimes.
Yan-Chao Li, Jia-Qing Xie, Ben Gao et al.
· 1 citation