Generalization of Diffusion Models Arises with a Balanced Representation Space
read the original abstract
Diffusion models excel at generating high-quality, diverse samples, yet they risk memorizing training data when overfit to the training objective. We analyze the distinctions between memorization and generalization in diffusion models through the lens of representation learning. By investigating a two-layer ReLU denoising autoencoder (DAE), we prove that (i) memorization corresponds to the model storing raw training samples in the learned weights for encoding and decoding, yielding localized spiky representations, whereas (ii) generalization arises when the model captures local data statistics, producing balanced representations. Furthermore, we validate these theoretical findings on real-world unconditional and text-to-image diffusion models, demonstrating that the same representation structures emerge in deep generative models with significant practical implications. Building on these insights, we propose a representation-based method for detecting memorization and a training-free editing technique that allows precise control via representation steering. Together, our results highlight that learning good representations is central to novel and meaningful generative modeling.
This paper has not been read by Pith yet.
Forward citations
Cited by 4 Pith papers
-
Proximal-Based Generative Modeling for Bayesian Inverse Problems
PGM replaces the intractable likelihood score in diffusion models with a closed-form Moreau score computed via proximal operators, enabling non-asymptotic sampling for inverse problems trained only on prior data.
-
A Theoretical Analysis of Memory and Overfitting Phenomena in Stochastic Interpolation Models
In the oracle continuous-time setting, stochastic interpolation models recover training samples exactly, with deviations controlled by discretization and estimation errors, leading to theoretical definitions of overfi...
-
Robust and Generalizable Safety Steering for Text-to-Image Diffusion Transformers
SafeDIG applies position-aware sparse feature transfer via SAEs in DiT models to reduce unsafe generations in target risk domains on FLUX.1 Dev and SD 3.5 while keeping source safety and quality.
-
Principles and Practice of Deep Representation Learning: or a Mathematical Theory of Memory
The book presents principles from optimization and information theory to explain deep network architectures and enable new interpretable models.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.