PulseAugur
EN
LIVE 19:11:18

Research links diffusion model memorization to local data coverage

By PulseAugur Editorial · [2 sources] ·

A new research paper proposes that memorization in diffusion models is not a global property but is instead governed by local data coverage. The study, which connects diffusion models to kernel density estimation, derives a theoretical criterion predicting memorization based on the density of training data in a model's neighborhood and the overall dataset size. This framework suggests that regions with low data coverage are prone to memorization, while dense regions facilitate generalization, a phenomenon validated empirically. AI

IMPACT Provides a local view of memorization in diffusion models, explaining occurrence based on data geometry.

RANK_REASON The cluster contains a research paper published on arXiv detailing theoretical and empirical findings about diffusion models.

Read on arXiv stat.ML →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

Research links diffusion model memorization to local data coverage

COVERAGE [2]

  1. arXiv stat.ML TIER_1 English(EN) · Claudia Merger, Sebastian Goldt ·

    Local Coverage Governs Memorization in Diffusion Models

    arXiv:2606.14390v1 Announce Type: cross Abstract: Memorization in diffusion models is often treated as a global property of the model or dataset. In practice, however, a single diffusion model can simultaneously generate both memorized and novel samples. Which training samples ar…

  2. arXiv stat.ML TIER_1 English(EN) · Sebastian Goldt ·

    Local Coverage Governs Memorization in Diffusion Models

    Memorization in diffusion models is often treated as a global property of the model or dataset. In practice, however, a single diffusion model can simultaneously generate both memorized and novel samples. Which training samples are most likely to be memorized? In this work, we sh…

RELATED ENTITIES

RELATED TOPICS