发表机构
RWTH Aachen University; University of Göttingen; TU Berlin(亚琛工业大学; 哥廷根大学; 柏林工业大学)
机构由 AI 辅助整理,请以论文原文为准。AI 中文总结
提出一种基于WFR JKO方案的重加权归一化流神经采样算法,证明其指数收敛性,并在多模态目标上验证了有效性。
AI 中文摘要
我们提出了一种用于从由未归一化玻尔兹曼密度指定的分布中采样的神经算法。我们的方法基于在Wasserstein--Fisher--Rao几何中针对Kullback--Leibler散度的Jordan--Kinderlehrer--Otto方案(WFR JKO方案)。我们的贡献有两方面。首先,我们证明了对于任意固定步长,精确的WFR JKO迭代在迭代次数趋于无穷时以指数速度收敛到目标分布。值得注意的是,这一结果不需要对目标分布作任何结构假设,如对数凹性或对数Sobolev不等式。其次,我们开发了WFR JKO方案的神经实现,该实现使用重加权归一化流对其传输和反应组件进行参数化。在具有挑战性的多模态目标上的数值实验证明了所提出方法的良好性能。
英文摘要
We propose a neural algorithm for sampling from distributions specified by unnormalized Boltzmann densities. Our approach is based on the Jordan--Kinderlehrer--Otto scheme for the Kullback--Leibler divergence in the Wasserstein--Fisher--Rao geometry (WFR JKO scheme). Our contributions are twofold. First, we prove that, for any fixed step size, the exact WFR JKO iterates converge exponentially fast to the target as the number of iterations tends to infinity. Notably, this result requires no structural assumptions on the target, such as log-concavity or a logarithmic Sobolev inequality. Second, we develop a neural implementation of the WFR JKO scheme that parametrizes its transport and reaction components using reweighted normalizing flows. Numerical experiments on challenging multimodal targets demonstrate the promising performance of the proposed method.