Memorization Control in Diffusion Models from Denoising-centric Perspective
This addresses the problem of memorization control in diffusion models for applications requiring generated data to closely match training distributions, representing an incremental improvement over existing approaches.
The paper tackled the problem of controlling memorization in diffusion models by analyzing it from a denoising-centric perspective, showing that uniform timestep sampling biases training toward memorization. They proposed a timestep sampling strategy that adjusts learning emphasis, which reduced memorization and improved distributional alignment in experiments on image and 1D signal generation tasks.
Controlling memorization in diffusion models is critical for applications that require generated data to closely match the training distribution. Existing approaches mainly focus on data centric or model centric modifications, treating the diffusion model as an isolated predictor. In this paper, we study memorization in diffusion models from a denoising centric perspective. We show that uniform timestep sampling leads to unequal learning contributions across denoising steps due to differences in signal to noise ratio, which biases training toward memorization. To address this, we propose a timestep sampling strategy that explicitly controls where learning occurs along the denoising trajectory. By adjusting the width of the confidence interval, our method provides direct control over the memorization generalization trade off. Experiments on image and 1D signal generation tasks demonstrate that shifting learning emphasis toward later denoising steps consistently reduces memorization and improves distributional alignment with training data, validating the generality and effectiveness of our approach.