Equal Path Cost, Unequal Output Effects: Understanding Perturbation Propagation in Diffusion Models

Key points

• Diffusion models have achieved remarkable success in generative modeling, with their sampling procedures routinely modified to control generation and improve efficiency.1
• These modifications introduce perturbations along the sampling trajectory, raising a central question: how do such perturbations affect generated output?1
• Within this framework, we quantify perturbation strength using the Kullback--Leibler (KL) divergence between perturbed and reference trajectory distributions, termed as path cost, which is shown to bound, but do not determine, changes in the output distribution.1
• We test our theoretical analysis through controlled interventions at equal path cost in pretrained diffusion models, revealing distinct patterns of output sensitivity across sampling stages and spatial frequencies.1

Sources (1)

• [1]Equal Path Cost, Unequal Output Effects: Understanding Perturbation Propagation in Diffusion Models
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 07:12 AM

  To address this question, we develop a theoretical framework to investigate perturbation propagation, combining dynamical analysis of the sampling process with an information-theoretic characterization of output responses.

  Diffusion models have achieved remarkable success in generative modeling, with their sampling procedures routinely modified to control generation and improve efficiency.

Extractive summary: sentences quoted from the sources.