arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2409.09650stat.MLcs.LG
收藏

生成式扩散模型中的条件采样

Conditional sampling within generative diffusion models

Linköping University(林雪平大学);Uppsala University(乌普萨拉大学)

Zheng Zhao, Ziwei Luo, Jens Sjölund, Thomas B. Schön

AI总结:

本文综述生成式扩散模型中的条件采样计算方法,梳理利用联合分布或带显式似然的预训练边缘分布构建条件采样器的关键技术。

摘要:

生成式扩散是一类功能强大的 Monte Carlo 采样器,它们利用桥接 Markov 过程来近似复杂的高维分布,例如图像处理和语言模型中出现的分布。尽管这些技术在上述领域取得了成功,但一个重要的开放挑战仍然存在:将这些技术扩展为从条件分布中采样,以满足例如 Bayesian 逆问题中的需求。在本文中,我们全面综述了生成式扩散模型中用于条件采样的现有计算方法。具体而言,我们重点介绍了两类关键方法:一类利用联合分布,另一类依赖具有显式似然的(预训练)边缘分布,用以构建条件生成式采样器。

英文摘要:

Generative diffusions are a powerful class of Monte Carlo samplers that leverage bridging Markov processes to approximate complex, high-dimensional distributions, such as those found in image processing and language models. Despite their success in these domains, an important open challenge remains: extending these techniques to sample from conditional distributions, as required in, for example, Bayesian inverse problems. In this paper, we present a comprehensive review of existing computational approaches to conditional sampling within generative diffusion models. Specifically, we highlight key methodologies that either utilise the joint distribution, or rely on (pre-trained) marginal distributions with explicit likelihoods, to construct conditional generative samplers.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑