New research links Föllmer processes to DDPMs, improving sampling efficiency
作者PulseAugur 编辑部·[13 个来源]·
Researchers have explored the connection between Föllmer processes and denoising diffusion probabilistic models (DDPMs), finding that discretizing Föllmer processes can yield optimal hyper-parameter settings for DDPM samplers. This approach has led to improved error bounds in terms of Wasserstein distance and KL divergence. Additionally, a new method called Forward-Learned Discrete Diffusion (FLDD) has been proposed, which learns the noising process to enable faster, few-step generation of high-quality samples.
AI
影响
Advances in diffusion model theory and sampling techniques could lead to more efficient and higher-quality generative AI.
排序理由
Multiple arXiv papers detailing theoretical advancements and new methods in diffusion models.
The Föllmer process is a Brownian motion conditioned to have a pre-specified distribution at time 1. This process can be interpreted as an "augmented" time-compressed version of the reverse stochastic differential equation (SDE) for the denoising diffusion probabilistic model (DD…
arXiv stat.ML
TIER_1English(EN)·Benedikt L\"utke Schwienhorst, Nadja Klein, Johannes Lederer·
arXiv:2605.22950v1 Announce Type: new Abstract: Score matching is an alternative to maximum likelihood estimation when the normalizing constant is unknown or too costly to evaluate. However, vanilla score matching has shown to be inefficient relative to maximum likelihood estimat…
arXiv stat.ML
TIER_1English(EN)·Samson Gourevitch, Yazid Janati, Dario Shariatian, Umut Simsekli, Eric Moulines, Eric P. Xing, Alain Durmus·
arXiv:2605.22765v1 Announce Type: cross Abstract: Discrete diffusion models are often trained through clean-data prediction, but the prediction can be used in different ways to define the reverse dynamics. In Masked Diffusion Models (MDM) these choices largely coincide, whereas i…
Score matching is an alternative to maximum likelihood estimation when the normalizing constant is unknown or too costly to evaluate. However, vanilla score matching has shown to be inefficient relative to maximum likelihood estimation for multimodal distributions with well-separ…
Discrete diffusion models are often trained through clean-data prediction, but the prediction can be used in different ways to define the reverse dynamics. In Masked Diffusion Models (MDM) these choices largely coincide, whereas in Uniform Diffusion Models (UDM) they do not. We s…
arXiv:2510.03824v2 Announce Type: replace-cross Abstract: The task of learning a diffusion-based neural sampler for drawing samples from an unnormalized target distribution can be viewed as a stochastic optimal control problem on path measures. However, the training of neural sam…
arXiv stat.ML
TIER_1English(EN)·Yifan Chen, Eric Vanden-Eijnden·
arXiv:2602.10989v2 Announce Type: replace-cross Abstract: We construct and analyze generative diffusions that transport a point mass to a prescribed target distribution over a finite time horizon using the stochastic interpolant framework. The drift is expressed as a conditional …
arXiv:2605.18040v1 Announce Type: new Abstract: The F\"ollmer process is a Brownian motion conditioned to have a pre-specified distribution at time 1. This process can be interpreted as an "augmented" time-compressed version of the reverse stochastic differential equation (SDE) f…
arXiv:2605.18069v1 Announce Type: new Abstract: This paper studies sampling error bounds for denoising diffusion probabilistic models (DDPMs) in the 2-Wasserstein distance. Our contributions are threefold. (i) Under general Lipschitz-type conditions on the score function and for …
arXiv stat.ML
TIER_1English(EN)·Grigory Bartosh, Teodora Pandeva, Sushrut Karmalkar, Javier Zazo·
arXiv:2605.18204v1 Announce Type: new Abstract: Discrete diffusion models are a powerful class of generative models with strong performance across many domains. For efficiency, however, discrete diffusion typically parameterizes the generative (reverse) process with factorized di…
Discrete diffusion models are a powerful class of generative models with strong performance across many domains. For efficiency, however, discrete diffusion typically parameterizes the generative (reverse) process with factorized distributions, which makes it difficult for the mo…
This paper studies sampling error bounds for denoising diffusion probabilistic models (DDPMs) in the 2-Wasserstein distance. Our contributions are threefold. (i) Under general Lipschitz-type conditions on the score function and for a broad class of variance schedules, including t…
The Föllmer process is a Brownian motion conditioned to have a pre-specified distribution at time 1. This process can be interpreted as an "augmented" time-compressed version of the reverse stochastic differential equation (SDE) for the denoising diffusion probabilistic model (DD…