Catastrophic Compositional Generation: Why Vanilla Diffusion Models Fail to Extrapolate

• URL: http://arxiv.org/abs/2606.23920v1
• Date: Mon, 22 Jun 2026 20:34:26 GMT
• Title: Catastrophic Compositional Generation: Why Vanilla Diffusion Models Fail to Extrapolate
• Abstract summary: We argue that no inference-time technique can efficiently produce samples from the target distribution in certain well-motivated settings.<n>Results show that score estimation error has a more catastrophic effect on performance when the target distribution is out-of-distribution with respect to the sources.
• Score: 16.740457845513358
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The task of compositional generation involves using a conditional generative model, trained only on a subset of the possible conditions, to produce samples from compositionally-defined target distributions such as a geometric combination of the source distributions. In this work, we argue that this task is often infeasible for vanilla conditional diffusion models: we conjecture that no inference-time technique can efficiently produce samples from the target distribution in certain well-motivated settings. This idea is supported by theory-guided generalization arguments and carefully-designed experiments on both synthetic and realistic data. In particular, while recent methods such as Feynman-Kac correction reduce inference-time approximation error, our results show that score estimation error has a more catastrophic effect on performance when the target distribution is out-of-distribution with respect to the sources, highlighting the need for a different approach to this task.

Related papers

• From Scores to Gibbs Correctors: Accelerating Uniform-Rate Discrete Diffusion Models [58.02657021435213]
  We propose a novel Gibbs-based corrector for discrete diffusion models, termed Gibbs-Accelerated Discrete Diffusion (GADD)<n>We show GADD achieves an overall sampling complexity of $mathcalO(mathrmpolylog (varepsilon-1)$, yielding the first such rate for diffusion-based samplers for uniform-rate discrete diffusion models.
  arXiv  Detail & Related papers  (2026-05-26T17:52:28Z)
• Error Propagation and Model Collapse in Diffusion Models: A Theoretical Study [27.894241484593735]
  Recursively training on synthetic data has been observed to significantly degrade performance in a wide range of tasks.<n>We theoretically analyze this phenomenon in the setting of score-based diffusion models.
  arXiv  Detail & Related papers  (2026-02-18T16:56:36Z)
• Adaptive Symmetrization of the KL Divergence [10.632997610787207]
  Many tasks in machine learning can be described as or reduced to learning a probability distribution given a finite set of samples.<n>A common approach is to minimize a statistical divergence between the (empirical) data distribution and a parameterized distribution, e.g., a normalizing flow (NF) or an energy-based model (EBM)
  arXiv  Detail & Related papers  (2025-11-14T10:41:59Z)
• Minimax Optimality of the Probability Flow ODE for Diffusion Models [8.15094483029656]
  This work develops the first end-to-end theoretical framework for deterministic ODE-based samplers.<n>We propose a smooth regularized score estimator that simultaneously controls both the $L2$ score error and the associated mean Jacobian error.<n>We demonstrate that the resulting sampler achieves the minimax rate in total variation distance, modulo logarithmic factors.
  arXiv  Detail & Related papers  (2025-03-12T17:51:29Z)
• Theory on Score-Mismatched Diffusion Models and Zero-Shot Conditional Samplers [49.97755400231656]
  We present the first performance guarantee with explicit dimensional dependencies for general score-mismatched diffusion samplers.<n>We show that score mismatches result in an distributional bias between the target and sampling distributions, proportional to the accumulated mismatch between the target and training distributions.<n>This result can be directly applied to zero-shot conditional samplers for any conditional model, irrespective of measurement noise.
  arXiv  Detail & Related papers  (2024-10-17T16:42:12Z)
• Constrained Diffusion Models via Dual Training [80.03953599062365]
  Diffusion processes are prone to generating samples that reflect biases in a training dataset. We develop constrained diffusion models by imposing diffusion constraints based on desired distributions. We show that our constrained diffusion models generate new data from a mixture data distribution that achieves the optimal trade-off among objective and constraints.
  arXiv  Detail & Related papers  (2024-08-27T14:25:42Z)
• Unveil Conditional Diffusion Models with Classifier-free Guidance: A Sharp Statistical Theory [87.00653989457834]
  Conditional diffusion models serve as the foundation of modern image synthesis and find extensive application in fields like computational biology and reinforcement learning. Despite the empirical success, theory of conditional diffusion models is largely missing. This paper bridges the gap by presenting a sharp statistical theory of distribution estimation using conditional diffusion models.
  arXiv  Detail & Related papers  (2024-03-18T17:08:24Z)
• Theoretical Insights for Diffusion Guidance: A Case Study for Gaussian Mixture Models [59.331993845831946]
  Diffusion models benefit from instillation of task-specific information into the score function to steer the sample generation towards desired properties. This paper provides the first theoretical study towards understanding the influence of guidance on diffusion models in the context of Gaussian mixture models.
  arXiv  Detail & Related papers  (2024-03-03T23:15:48Z)
• Flow Away your Differences: Conditional Normalizing Flows as an Improvement to Reweighting [0.0]
  We present an alternative to reweighting techniques for modifying distributions to account for a desired change in an underlying conditional distribution. We employ conditional normalizing flows to learn the full conditional probability distribution. In our examples, this leads to a statistical precision up to three times greater than using reweighting techniques with identical sample sizes for the source and target distributions.
  arXiv  Detail & Related papers  (2023-04-28T16:33:50Z)
• Bi-Noising Diffusion: Towards Conditional Diffusion Models with Generative Restoration Priors [64.24948495708337]
  We introduce a new method that brings predicted samples to the training data manifold using a pretrained unconditional diffusion model. We perform comprehensive experiments to demonstrate the effectiveness of our approach on super-resolution, colorization, turbulence removal, and image-deraining tasks.
  arXiv  Detail & Related papers  (2022-12-14T17:26:35Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.