Entropic Time Schedulers for Generative Diffusion Models
This work addresses a key bottleneck in diffusion model efficiency for practitioners by optimizing inference performance without additional computational cost, though it is incremental as it builds on existing scheduling methods.
The paper tackles the problem of noise scheduling in generative diffusion models by introducing an entropic time scheduler that selects sampling points based on entropy to ensure equal information contribution, and it shows that this approach substantially improves image quality in pretrained EDM2 models on ImageNet, with FID and FD-DINO scores increasing without extra function evaluations, especially in low NFE regimes.
The practical performance of generative diffusion models depends on the appropriate choice of the noise scheduling function, which can also be equivalently expressed as a time reparameterization. In this paper, we present a time scheduler that selects sampling points based on entropy rather than uniform time spacing, ensuring that each point contributes an equal amount of information to the final generation. We prove that this time reparameterization does not depend on the initial choice of time. Furthermore, we provide a tractable exact formula to estimate this \emph{entropic time} for a trained model using the training loss without substantial overhead. Alongside the entropic time, inspired by the optimality results, we introduce a rescaled entropic time. In our experiments with mixtures of Gaussian distributions and ImageNet, we show that using the (rescaled) entropic times greatly improves the inference performance of trained models. In particular, we found that the image quality in pretrained EDM2 models, as evaluated by FID and FD-DINO scores, can be substantially increased by the rescaled entropic time reparameterization without increasing the number of function evaluations, with greater improvements in the few NFEs regime.