H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology(哈里·米尔顿·斯图尔特工业与系统工程学院,佐治亚理工学院);Department of Mathematics, Duke University(杜克大学数学系)
Flow Matching (FM) is a simulation-free method for learning a continuous, invertible flow that interpolates between two distributions, and in particular generates data from noise. Inspired by the variational nature of the diffusion process as a gradient flow, we introduce a stepwise FM model, Local Flow Matching (LFM), which sequentially learns a sequence of FM submodels, each matching a diffusion process up to the time-step size in the data-to-noise direction. In each step, the two distributions to be interpolated by the sub-flow model are closer than those in the full-flow matching model, which interpolates data to noise distributions, enabling smaller models with more efficient training. This variational perspective also allows us to prove a theoretical generation guarantee for the proposed flow model in terms of the $χ^2$-divergence between the generated and true data distributions, leveraging the contraction property of the diffusion process. In practice, the stepwise structure of LFM is naturally amenable to model distillation, and various distillation techniques can be applied to accelerate generation. We empirically demonstrate that LFM achieves competitive generative performance compared to FM on unconditional generation of tabular and image datasets, and on conditional generation of robotic manipulation policies.
University of Toronto(多伦多大学);Canadian Institute for Advanced Research(加拿大高等研究院)
我们考虑从数据中学习深度生成模型的问题。我们制定了一种方法,该方法通过多层感知器的单次前馈来生成独立样本,如最近提出的生成对抗网络(Goodfellow et al.,2014年)。然而,训练一个生成对抗网络需要仔细优化一个困难的极小极大程序。相反,我们利用了一种被称为最大平均差异(MMD)的统计假设检验技术,这导致了一个简单的目标,可以被解释为匹配数据集和模型样本之间的所有统计阶数,并且可以通过反向传播进行训练。我们通过将我们的生成网络与自动编码器网络相结合,使用MMD来学习生成代码,然后可以解码生成样本,从而进一步提高了这种方法的性能。我们表明,这些技术的组合产生优秀的生成模型相比,MNIST和多伦多人脸数据库上测量的基线方法。
Emerging Convolutions for Generative Normalizing Flows
University of Amsterdam(阿姆斯特丹大学);Canadian Institute for Advanced Research (CIFAR)(加拿大高等研究院)
生成流是有吸引力的,因为它们承认精确的似然优化和有效的图像合成。最近,Kingma & Dhariwal(2018)用Glow证明了生成流能够生成高质量的图像。我们将Glow中提出的1 x 1卷积推广到可逆的d x d卷积,因为它们在通道和空间轴上操作,所以更灵活。我们提出了两种方法来产生具有与标准卷积相同的感受野的可逆卷积:通过链接特定的自回归卷积来获得新兴卷积,并且在频域中解耦周期卷积。我们的实验表明,d x d卷积的灵活性显着提高了星系图像、CIFAR10和ImageNet上生成流模型的性能。
Generative Modeling of Quantum Distribution with Functional Flow Matching
Graduate School of Data Science, Seoul National University, Seoul, Korea(首尔国立大学数据科学研究生院);Department of Statistics and Data Science, Yonsei University, Seoul, Korea(延世大学统计与数据科学系);Department of Applied Statistics, Yonsei University, Seoul, Korea(延世大学应用统计系);Google Research, Mountain View, California, United States(谷歌研究)
A Theoretical Analysis of Discrete Flow Matching Generative Models
Center for Foundation Models and Generative AI, Northwestern University, Evanston, IL 60208, USA(基础模型与生成式人工智能中心,西北大学,伊利诺伊州埃文斯顿,60208,美国);Department of Computer Science, Northwestern University, Evanston, IL 60208, USA(计算机科学系,西北大学,伊利诺伊州埃文斯顿,60208,美国);University of California, Berkeley, Berkeley, CA 94720, USA(加州大学伯克利分校,伯克利,加利福尼亚州,94720,美国);Ensemble AI, San Francisco, CA 94133, USA(Ensemble AI,旧金山,加利福尼亚州,94133,美国);Department of Statistics and Data Science, Northwestern University, Evanston, IL 60208, USA(统计与数据科学系,西北大学,伊利诺伊州埃文斯顿,60208,美国)
University of North Carolina Charlotte(北卡罗来纳大学夏洛特分校);Montreal Institute for Learning Algorithms(蒙特利尔学习算法研究所);Carnegie Mellon University(卡内基梅隆大学);University of Maryland College Park(马里兰大学学院公园分校);Korea University(高丽大学);IIIT Delhi(德里印度理工学院);Electronics and Telecommunications Research Institute(电子与电信研究院)
Flow matching-based generative models for MIMO channel estimation
State Key Laboratory of Networking and Switching Technology(网络与交换技术国家重点实验室);Beijing University of Posts and Telecommunications(北京邮电大学);ZGC Institute of Ubiquitous-X Innovation and Applications(ZGC 全息-X创新与应用研究院)
Force Matching with Relativistic Constraints: A Physics-Inspired Approach to Stable and Efficient Generative Modeling
Wyoming Seminary(怀俄明神学院);Middle Tennessee State University(中田纳西州立大学);Stevens Institute of Technology(斯蒂文斯理工学院);The University of Hong Kong(香港大学);University of Wisconsin-Madison(威斯康星大学麦迪逊分校);Tsinghua University(清华大学);Simons Institute for the Theory of Computing, University of California, Berkeley(加州大学伯克利分校西蒙斯理论计算研究所);Anhui University(安徽大学)
Theoretical Guarantees for High Order Trajectory Refinement in Generative Flows
The University of Texas at Austin(德克萨斯大学奥斯汀分校);Stevens Institute of Technology(史蒂文斯理工学院);The University of Hong Kong(香港大学);University of Wisconsin-Madison(威斯康星大学麦迪逊分校);South China University of Technology(华南理工大学);The Simons Institute for the Theory of Computing(西蒙斯理论计算研究所)