aiqbench analysis
The paper proposes "model information," a tractable measure of how much a diffusion model's outputs depend on its specific training set, and uses it to identify two scaling regimes: a nonlinear transition from memorization to generalization, then a linear regime of improving sample quality. It also fits a scaling law that predicts memorization and sample quality and reports that it extrapolates beyond the fitted range. Readers training or evaluating generative models should note that these are results from a single arXiv preprint, so the extrapolation claims need independent replication across architectures and datasets before they can guide planning.
Information-Theoretic Scaling Laws for Generative Diffusion Training
arXiv:2610.33337v1 Announce Type: new Abstract: Diffusion models trained on increasingly large datasets sampled from the same data distribution will converge to the same learned distribution. However, in practice, a model trained on finite data depends on the particular training dataset; this dependence is reflected in properties of the model's outputs, such as memorization and sample quality. To understand this dependence through an information-theoretic lens, we study \textit{model information}, a computationally tractable measure of how much samples from the trained model depend on the particular training dataset. We explore its scaling behavior as model size and training dataset size vary. Specifically, we uncover two qualitatively distinct scaling regimes: (1) a nonlinear model information regime marking the transition from pure memorization to imperfect generalization, and (2) a generalization regime characterized by linear scaling of the model information and progressively improved generalization and sample quality. We then posit an information-theoretically principled scaling law to quantify these two regimes, and show that this law yields induced scaling laws for memorization and sample quality. We finally demonstrate that these induced scaling laws extrapolate well outside the fit's domain. Our results demonstrate information-theoretic principles for memorization and sample quality scaling laws in diffusion models. Our code is publicly available on \href{https://github.com/DruvPai/diffusion_info_theory_scaling_law}{GitHub}.
Abstract:Diffusion models trained on increasingly large datasets sampled from the same data distribution will converge to the same learned distribution. However, in practice, a model trained on finite data depends on the particular training dataset; this dependence is reflected in properties of the model's outputs, such as memorization and sample quality. To understand this dependence through an information-theoretic lens, we study \textit{model information}, a computationally tractable measure of how much samples from the trained model depend on the particular training dataset. We explore its scaling behavior as model size and training dataset size vary. Specifically, we uncover two qualitatively distinct scaling regimes: (1) a nonlinear model information regime marking the transition from pure memorization to imperfect generalization, and (2) a generalization regime characterized by linear scaling of the model information and progressively improved generalization and sample quality. We then posit an information-theoretically principled scaling law to quantify these two regimes, and show that this law yields induced scaling laws for memorization and sample quality. We finally demonstrate that these induced scaling laws extrapolate well outside the fit's domain. Our results demonstrate information-theoretic principles for memorization and sample quality scaling laws in diffusion models. Our code is publicly available on \href{this https URL}{GitHub}.
Machine Learning (cs.LG)
arXiv:2610.33337 [cs.LG]
(or arXiv:2610.33337v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.33337 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 296 words