analiza aiqbench
A new arXiv preprint proposes masking most training pixels in diffusion models to reduce spatially inconsistent outputs and memorization on small datasets. The authors report that the method keeps competitive FID scores even at 98% masking, though this is a preprint whose results have not been peer-reviewed. Readers should check whether the gains hold on larger, standard benchmarks and whether the simplicity of the masking approach carries over to other generative setups.
Sparsely Supervised Diffusion
arXiv:2602.02699v2 Announce Type: replace Abstract: Diffusion models have shown remarkable success across a wide range of generative tasks. However, they often suffer from spatially inconsistent generation, arguably due to excessive correlations learned by the model. This can produce samples that are locally plausible but globally inconsistent. We propose a principled method to mitigate this issue. Our method, sparsely supervised diffusion (SSD), is a simple yet effective masking framework that can be implemented with only a few lines of code. We analytically show that SSD fundamentally alters diffusion model training by modifying the spectrum of the data covariance and that it suppresses correlations in the data covariance matrix. Experiments show that our method more accurately approximates the underlying population score function, reduces memorization on small datasets, and promotes the use of essential contextual information during generation. Moreover, even when up to 98\% of pixels are masked, it achieves competitive FID scores across a range of datasets and, importantly, avoids the training instability commonly observed on small datasets.
Abstract:Diffusion models have shown remarkable success across a wide range of generative tasks. However, they often suffer from spatially inconsistent generation, arguably due to excessive correlations learned by the model. This can produce samples that are locally plausible but globally inconsistent. We propose a principled method to mitigate this issue. Our method, sparsely supervised diffusion (SSD), is a simple yet effective masking framework that can be implemented with only a few lines of code. We analytically show that SSD fundamentally alters diffusion model training by modifying the spectrum of the data covariance and that it suppresses correlations in the data covariance matrix. Experiments show that our method more accurately approximates the underlying population score function, reduces memorization on small datasets, and promotes the use of essential contextual information during generation. Moreover, even when up to 98\% of pixels are masked, it achieves competitive FID scores across a range of datasets and, importantly, avoids the training instability commonly observed on small datasets.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2602.02699 [cs.LG]
(or arXiv:2602.02699v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2602.02699 Focus to learn more arXiv-issued DOI via DataCite
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.
Text integral sub licența deschisă a publicației · 256 cuvinte
Subiecte conexe
Evaluating the Representation Space of Diffusion Models via Self-Supervised Principles
arXiv · 9 iun. 2026
ISS Policy: Scalable Diffusion Policy with Implicit Scene Supervision
arXiv · 16 sept. 2026
Entropy as a Structural Prior: How a Log-Barrier on DiT Belief Space Drives Musical Diversity and Development
arXiv · 8 iun. 2026
Revisiting Spectral Representations in Generative Diffusion Models
arXiv · 9 sept. 2026