Diagnosing and Improving Diffusion Models by Estimating the Optimal Loss Value
This work addresses a diagnostic and improvement challenge for diffusion models in generative AI, offering incremental but practical tools for researchers and practitioners.
The authors tackled the problem that diffusion models' loss values are not indicative of data-fitting quality because the optimal loss is unknown, and they derived a closed-form optimal loss and estimators to diagnose training and improve performance, showing better scaling law demonstration with models up to 1.5B parameters.
Diffusion models have achieved remarkable success in generative modeling. Despite more stable training, the loss of diffusion models is not indicative of absolute data-fitting quality, since its optimal value is typically not zero but unknown, leading to confusion between large optimal loss and insufficient model capacity. In this work, we advocate the need to estimate the optimal loss value for diagnosing and improving diffusion models. We first derive the optimal loss in closed form under a unified formulation of diffusion models, and develop effective estimators for it, including a stochastic variant scalable to large datasets with proper control of variance and bias. With this tool, we unlock the inherent metric for diagnosing the training quality of mainstream diffusion model variants, and develop a more performant training schedule based on the optimal loss. Moreover, using models with 120M to 1.5B parameters, we find that the power law is better demonstrated after subtracting the optimal loss from the actual training loss, suggesting a more principled setting for investigating the scaling law for diffusion models.