The Interplay of Data Structure and Imbalance in the Learning Dynamics of Diffusion Models

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Nicoletti, Flavio, Ma, Chenxiao, Ventura, Enrico, Saglietti, Luca, Mannelli, Stefano Sarao
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915989570977792
author Nicoletti, Flavio
Ma, Chenxiao
Ventura, Enrico
Saglietti, Luca
Mannelli, Stefano Sarao
author_facet Nicoletti, Flavio
Ma, Chenxiao
Ventura, Enrico
Saglietti, Luca
Mannelli, Stefano Sarao
contents Real-world datasets are inherently heterogeneous, yet how per-class structural differences and sampling imbalance shape the training dynamics of diffusion models-and potentially exacerbate disparities-remains poorly understood. While models typically transition from an initial phase of generalization to memorizing the training set, existing theory assumes homogeneous data, leaving open how class imbalance and heterogeneity reshape these dynamics. In this work, we develop a high-dimensional analytical framework to study class-dependent learning in score-based diffusion models. Analyzing a random-features model trained on Gaussian mixtures, we derive the feature-covariance spectrum to characterize per-class generalization and memorization times. We reveal the explicit hierarchy governing these dynamics: class variance is the primary determinant of learning order-consistently favoring higher-variance classes-while centroid geometry plays a secondary role. Sampling imbalance acts as a modulator that can reverse this ordering and, under strong imbalance, forces minority classes to acquire distinct, delayed speciation times during backward diffusion. Together, these results suggest that diffusion models can memorize some classes while others remain insufficiently learned. We validate our theoretical predictions empirically using U-Net models trained on Fashion MNIST.
format Preprint
id arxiv_https___arxiv_org_abs_2605_06367
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle The Interplay of Data Structure and Imbalance in the Learning Dynamics of Diffusion Models
Nicoletti, Flavio
Ma, Chenxiao
Ventura, Enrico
Saglietti, Luca
Mannelli, Stefano Sarao
Machine Learning
Disordered Systems and Neural Networks
Real-world datasets are inherently heterogeneous, yet how per-class structural differences and sampling imbalance shape the training dynamics of diffusion models-and potentially exacerbate disparities-remains poorly understood. While models typically transition from an initial phase of generalization to memorizing the training set, existing theory assumes homogeneous data, leaving open how class imbalance and heterogeneity reshape these dynamics. In this work, we develop a high-dimensional analytical framework to study class-dependent learning in score-based diffusion models. Analyzing a random-features model trained on Gaussian mixtures, we derive the feature-covariance spectrum to characterize per-class generalization and memorization times. We reveal the explicit hierarchy governing these dynamics: class variance is the primary determinant of learning order-consistently favoring higher-variance classes-while centroid geometry plays a secondary role. Sampling imbalance acts as a modulator that can reverse this ordering and, under strong imbalance, forces minority classes to acquire distinct, delayed speciation times during backward diffusion. Together, these results suggest that diffusion models can memorize some classes while others remain insufficiently learned. We validate our theoretical predictions empirically using U-Net models trained on Fashion MNIST.
title The Interplay of Data Structure and Imbalance in the Learning Dynamics of Diffusion Models
topic Machine Learning
Disordered Systems and Neural Networks
url https://arxiv.org/abs/2605.06367