Can We Change the Stroke Size for Easier Diffusion?
For researchers working on diffusion models, this is an incremental idea that modifies training targets to improve performance in low signal-to-noise regimes.
The paper proposes controlling stroke size (effective roughness of predictions) in diffusion models to ease the low signal-to-noise challenge, analyzing trade-offs theoretically and empirically.
Diffusion models can be challenged in the low signal-to-noise regime, where they have to make pixel-level predictions despite the presence of high noise. The geometric intuition is akin to using the finest stroke for oil painting throughout, which may be ineffective. We therefore study stroke-size control as a controlled intervention that changes the effective roughness of the supervised target, predictions and perturbations across timesteps, in an attempt to ease the low signal-to-noise challenge. We analyze the advantages and trade-offs of the intervention both theoretically and empirically. Code will be released.