From Modes to Memories: Characterizing the Scale-Space Dynamics of Diffusion Models

2026-09-30

AI Summary

This study addresses the memorization phenomenon in diffusion models, which frequently arises from overfitting or data duplication yet remains difficult to quantify and attribute precisely using existing methods. By modeling the denoising process as a deterministic dynamical system at fixed noise scales, this work proposes a geometric analysis framework grounded in scale-space dynamics. Specifically, it introduces a critical noise scale, σc, to characterize sample persistence under multi-scale density smoothing, thereby constructing an interpretable memorization metric. Applied to Stable Diffusion, the proposed approach effectively identifies both fully and partially memorized samples while revealing their distributional properties in image space and text-dependent characteristics. Ultimately, this research establishes a novel paradigm for training data attribution in conditional generative models.

Abstract

Diffusion models are typically viewed as stochastic processes that transform noise into data. We take a complementary perspective: a diffusion model defines a family of deterministic dynamical systems indexed by noise scale. At each fixed scale $σ$, we treat the denoiser as a self-map and study its dynamics. For an exact denoiser, fixed points correspond to critical points of the smoothed data density, while attractors correspond to its modes; as $σ$ increases, sample-level modes merge into progressively coarser ones. This suggests a geometric view of memorization: examples that receive excess probability mass due to duplication or overfitting, as well as outliers, should remain distinguishable under stronger smoothing than ordinary examples. We quantify this persistence by the critical scale $σ_c$, the largest noise scale at which an example is retained by the fixed-scale dynamics. In conditional models, the same construction extends naturally to image--caption pairs. Experiments in controlled settings and on large-scale models show that $σ_c$ tracks memorization arising from duplication, overfitting, and outliers, and identifies both memorized and partially memorized examples in Stable Diffusion. Moreover, $σ_c$ yields interpretable measures of the image spatial distribution and caption dependence of memorization.

Problem

• Diffusion Models
• Memorization
• Scale-Space Dynamics
• Overfitting

Innovation

• Diffusion Models
• Scale-Space Dynamics
• Memorization
• Critical Scale
• Deterministic Dynamical Systems