Understanding Sampler Stochasticity in Training Diffusion Models for RLHF

• URL: http://arxiv.org/abs/2510.10767v1
• Date: Sun, 12 Oct 2025 19:08:38 GMT
• Title: Understanding Sampler Stochasticity in Training Diffusion Models for RLHF
• Authors: Jiayuan Sheng, Hanyang Zhao, Haoxian Chen, David D. Yao, Wenpin Tang,
• Abstract summary: This paper theoretically characterizes the reward gap and provides non-vacuous bounds for general diffusion models.<n> Empirically, our findings through large-scale experiments on text-to-image models validate that reward gaps consistently narrow over training.
• Score: 11.537564997052606
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Reinforcement Learning from Human Feedback (RLHF) is increasingly used to fine-tune diffusion models, but a key challenge arises from the mismatch between stochastic samplers used during training and deterministic samplers used during inference. In practice, models are fine-tuned using stochastic SDE samplers to encourage exploration, while inference typically relies on deterministic ODE samplers for efficiency and stability. This discrepancy induces a reward gap, raising concerns about whether high-quality outputs can be expected during inference. In this paper, we theoretically characterize this reward gap and provide non-vacuous bounds for general diffusion models, along with sharper convergence rates for Variance Exploding (VE) and Variance Preserving (VP) Gaussian models. Methodologically, we adopt the generalized denoising diffusion implicit models (gDDIM) framework to support arbitrarily high levels of stochasticity, preserving data marginals throughout. Empirically, our findings through large-scale experiments on text-to-image models using denoising diffusion policy optimization (DDPO) and mixed group relative policy optimization (MixGRPO) validate that reward gaps consistently narrow over training, and ODE sampling quality improves when models are updated using higher-stochasticity SDE training.

Related papers

• G$^2$RPO: Granular GRPO for Precise Reward in Flow Models [74.21206048155669]
  We propose a novel Granular-GRPO (G$2$RPO) framework that achieves precise and comprehensive reward assessments of sampling directions.<n>We introduce a Multi-Granularity Advantage Integration module that aggregates advantages computed at multiple diffusion scales.<n>Our G$2$RPO significantly outperforms existing flow-based GRPO baselines.
  arXiv  Detail & Related papers  (2025-10-02T12:57:12Z)
• Inference-Time Scaling of Diffusion Language Models with Particle Gibbs Sampling [70.8832906871441]
  We study how to steer generation toward desired rewards without retraining the models.<n>Prior methods typically resample or filter within a single denoising trajectory, optimizing rewards step-by-step without trajectory-level refinement.<n>We introduce particle Gibbs sampling for diffusion language models (PG-DLM), a novel inference-time algorithm enabling trajectory-level refinement while preserving generation perplexity.
  arXiv  Detail & Related papers  (2025-07-11T08:00:47Z)
• Diffusion Sampling Path Tells More: An Efficient Plug-and-Play Strategy for Sample Filtering [18.543769006014383]
  Diffusion models often exhibit inconsistent sample quality due to variations inherent in their sampling trajectories.<n>We introduce CFG-Rejection, an efficient, plug-and-play strategy that filters low-quality samples at an early stage of the denoising process.<n>We validate the effectiveness of CFG-Rejection in image generation through extensive experiments.
  arXiv  Detail & Related papers  (2025-05-29T11:08:24Z)
• Training-free Diffusion Model Alignment with Sampling Demons [15.400553977713914]
  We propose an optimization approach, dubbed Demon, to guide the denoising process at inference time without backpropagation through reward functions or model retraining.<n>Our approach works by controlling noise distribution in denoising steps to concentrate density on regions corresponding to high rewards through optimization.<n>Our experiments show that the proposed approach significantly improves the average aesthetics scores text-to-image generation.
  arXiv  Detail & Related papers  (2024-10-08T07:33:49Z)
• Semi-Implicit Denoising Diffusion Models (SIDDMs) [50.30163684539586]
  Existing models such as Denoising Diffusion Probabilistic Models (DDPM) deliver high-quality, diverse samples but are slowed by an inherently high number of iterative steps. We introduce a novel approach that tackles the problem by matching implicit and explicit factors. We demonstrate that our proposed method obtains comparable generative performance to diffusion-based models and vastly superior results to models with a small number of sampling steps.
  arXiv  Detail & Related papers  (2023-06-21T18:49:22Z)
• A Geometric Perspective on Diffusion Models [57.27857591493788]
  We inspect the ODE-based sampling of a popular variance-exploding SDE. We establish a theoretical relationship between the optimal ODE-based sampling and the classic mean-shift (mode-seeking) algorithm.
  arXiv  Detail & Related papers  (2023-05-31T15:33:16Z)
• Reflected Diffusion Models [93.26107023470979]
  We present Reflected Diffusion Models, which reverse a reflected differential equation evolving on the support of the data. Our approach learns the score function through a generalized score matching loss and extends key components of standard diffusion models.
  arXiv  Detail & Related papers  (2023-04-10T17:54:38Z)
• Fast Inference in Denoising Diffusion Models via MMD Finetuning [23.779985842891705]
  We present MMD-DDM, a novel method for fast sampling of diffusion models. Our approach is based on the idea of using the Maximum Mean Discrepancy (MMD) to finetune the learned distribution with a given budget of timesteps. Our findings show that the proposed method is able to produce high-quality samples in a fraction of the time required by widely-used diffusion models.
  arXiv  Detail & Related papers  (2023-01-19T09:48:07Z)
• How Much is Enough? A Study on Diffusion Times in Score-based Generative Models [76.76860707897413]
  Current best practice advocates for a large T to ensure that the forward dynamics brings the diffusion sufficiently close to a known and simple noise distribution. We show how an auxiliary model can be used to bridge the gap between the ideal and the simulated forward dynamics, followed by a standard reverse diffusion process.
  arXiv  Detail & Related papers  (2022-06-10T15:09:46Z)

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.