arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2409.14993cs.AIcs.CV
收藏

多模态生成式AI:多模态大语言模型、扩散模型与统一

Multi-modal Generative AI: Multi-modal LLMs, Diffusions, and the Unification

Department of Computer Science, Beijing Information Science and Technology National Research Center, Tsinghua University(计算机系,北京信息科学与技术国家研究中心,清华大学)

Xin Wang, Yuwei Zhou, Bin Huang, Hong Chen, Wenwu Zhu

AI总结:

本文综述了多模态生成式AI,涵盖多模态LLMs、扩散模型及统一模型的设计与应用,探讨了统一理解和生成的方法及未来研究方向。

摘要:

多模态生成式人工智能(人工智能)已吸引了学术界和工业界越来越多的关注。特别是,两种主导的技术家族已出现:i)多模态大语言模型(LLMs)展示了在多模态理解方面的出色能力;ii)扩散模型在多模态生成方面表现出显著的多模态能力。因此,本文全面概述了多模态生成式人工智能,包括多模态LLMs、扩散模型以及理解和生成的统一。为了为统一模型奠定坚实的基础,我们首先分别对多模态LLMs和扩散模型进行了详细的回顾,包括其概率建模过程、多模态架构设计以及先进的应用,如图像/视频LLMs以及文本到图像/视频生成。此外,我们探索了朝着理解和生成统一模型的新兴努力。为了实现理解和生成的统一,我们调查了关键设计,包括基于自回归和扩散的建模,以及密集和专家混合(MoE)架构。我们然后介绍了几种统一模型的策略,分析了它们的潜在优势和劣势。此外,我们总结了广泛用于多模态生成式人工智能预训练的常用数据集。最后但同样重要的是,我们提出了几种具有挑战性的未来研究方向,这些方向可能有助于多模态生成式人工智能的持续发展。

英文摘要:

Multi-modal generative AI (Artificial Intelligence) has attracted increasing attention from both academia and industry. Particularly, two dominant families of techniques have emerged: i) Multi-modal large language models (LLMs) demonstrate impressive ability for multi-modal understanding; and ii) Diffusion models exhibit remarkable multi-modal powers in terms of multi-modal generation. Therefore, this paper provides a comprehensive overview of multi-modal generative AI, including multi-modal LLMs, diffusions, and the unification for understanding and generation. To lay a solid foundation for unified models, we first provide a detailed review of both multi-modal LLMs and diffusion models respectively, including their probabilistic modeling procedure, multi-modal architecture design, and advanced applications to image/video LLMs as well as text-to-image/video generation. Furthermore, we explore the emerging efforts toward unified models for understanding and generation. To achieve the unification of understanding and generation, we investigate key designs including autoregressive-based and diffusion-based modeling, as well as dense and Mixture-of-Experts (MoE) architectures. We then introduce several strategies for unified models, analyzing their potential advantages and disadvantages. In addition, we summarize the common datasets widely used for multi-modal generative AI pretraining. Last but not least, we present several challenging future research directions which may contribute to the ongoing advancement of multi-modal generative AI.

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。

↑