aiqbench analysis
This paper is clearly about AI: a method for training label-conditional diffusion models when class labels are noisy, ambiguous, or missing. The core idea is to treat the true label as a latent variable and jointly learn the clean data distribution and label posteriors, with a timestep sampling trick to keep training tractable. The claims of outperforming specialized baselines are based on synthetic and real-world benchmarks, so readers should check how much the gains depend on the specific noise models tested and on the assumed prior knowledge. For practitioners training generative models on imperfectly labeled data, the practical question is whether DELTA's robustness holds on messier, real-world annotation pipelines.
DELTA: Robustly Training Label-Conditional Diffusion Models with Weak Annotations
arXiv:2510.03016v3 Announce Type: replace Abstract: While label-conditional diffusion models have recently exhibited remarkable generative capabilities, their success heavily relies on massive, cleanly labeled datasets. In practice, categorical supervision is rarely perfect: it is often corrupted by noise, clouded by ambiguity, or partially missing. Training directly on such weak annotations severely degrades generation quality, and existing robust methods offer only fragmented solutions that depend on scarce auxiliary priors. To overcome these limitations, we propose DELTA, a unified framework that robustly trains Diffusion modELs across all three weak annoTAtion types. By treating the unknown true label as a latent variable, DELTA optimizes a principled variational objective that jointly recovers the clean data distribution and infers the true label posteriors. To make this joint training computationally tractable, we further introduce a median-centered timestep sampling strategy that efficiently concentrates estimation where the diffusion process is most informative. Extensive experiments on synthetic and real-world benchmarks demonstrate that DELTA produces high-fidelity, class-consistent samples across diverse weak supervision scenarios, outperforming specialized baselines while demanding strictly less prior knowledge.
Abstract:While label-conditional diffusion models have recently exhibited remarkable generative capabilities, their success heavily relies on massive, cleanly labeled datasets. In practice, categorical supervision is rarely perfect: it is often corrupted by noise, clouded by ambiguity, or partially missing. Training directly on such weak annotations severely degrades generation quality, and existing robust methods offer only fragmented solutions that depend on scarce auxiliary priors. To overcome these limitations, we propose DELTA, a unified framework that robustly trains Diffusion modELs across all three weak annoTAtion types. By treating the unknown true label as a latent variable, DELTA optimizes a principled variational objective that jointly recovers the clean data distribution and infers the true label posteriors. To make this joint training computationally tractable, we further introduce a median-centered timestep sampling strategy that efficiently concentrates estimation where the diffusion process is most informative. Extensive experiments on synthetic and real-world benchmarks demonstrate that DELTA produces high-fidelity, class-consistent samples across diverse weak supervision scenarios, outperforming specialized baselines while demanding strictly less prior knowledge.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2510.03016 [cs.LG]
(or arXiv:2510.03016v3 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2510.03016 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.
Full text under the publisher's open licence · 261 words