High-dimensional density estimation with tensorizing flow

DOI10.1007/S40687-023-00395-XMaRDI QIDQ6173686zbMATH OpenOpenAlexFDO

Authors Yinuo Ren, Hongli Zhao, Yuehaw Khoo, Lexing Ying

Publication date 21 July 2023

Published in Research in the Mathematical Sciences (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/2212.00759

zbMATH Keywords

density estimationflow-based generative modelingmaximum likelihood estimationtensor-train

Mathematics Subject Classification ID

Statistics (62-XX) Fluid mechanics (76-XX)

Abstract: We propose the tensorizing flow method for estimating high-dimensional probability density functions from the observed data. The method is based on tensor-train and flow-based generative modeling. Our method first efficiently constructs an approximate density in the tensor-train form via solving the tensor cores from a linear system based on the kernel density estimators of low-dimensional marginals. We then train a continuous-time flow model from this tensor-train density to the observed empirical distribution by performing a maximum likelihood estimation. The proposed method combines the optimization-less feature of the tensor-train with the flexibility of the flow-based generative models. Numerical results are included to demonstrate the performance of the proposed method.

Cites work

• 10.1162/153244303322533223
• A Multilinear Singular Value Decomposition
• An Introduction to Fluid Dynamics
• An introduction to the Ginzburg-Landau theory of phase transitions and nonequilibrium patterns
• Approximation and sampling of multivariate probability distributions in the tensor train decomposition
• Computation of extreme eigenvalues in higher dimensions using block tensor train format
• Density estimation by dual ascent of the log-likelihood
• Duality of graphical models and tensor networks
• Exponential decay of correlations implies area law
• Generative adversarial networks are special cases of artificial curiosity (1990) and also closely related to predictability minimization (1991)
• Generative modeling via tree tensor network states
• Hierarchical Singular Value Decomposition of Tensors
• scientific article; zbMATH DE number 6377992 (Why is no real title available?)
• scientific article; zbMATH DE number 193258 (Why is no real title available?)
• scientific article; zbMATH DE number 2171466 (Why is no real title available?)
• scientific article; zbMATH DE number 7626702 (Why is no real title available?)
• scientific article; zbMATH DE number 3344105 (Why is no real title available?)
• Matrix product state representations
• Minimum action method for the study of rare events
• On low-rank approximability of solutions to high-dimensional operator equations and eigenvalue problems
• On the Best Rank-1 and Rank-(R₁ ,R₂ ,. . .,R_N) Approximation of Higher-Order Tensors
• Parallel Algorithms for Computing the Tensor-Train Decomposition
• Pattern recognition and machine learning.
• Riemannian optimization for high-dimensional tensor completion
• Tensor networks and hierarchical tensors for the solution of high-dimensional partial differential equations
• Tensor Train Neighborhood Preserving Embedding
• Tensor-train decomposition
• Training Products of Experts by Minimizing Contrastive Divergence
• TT-cross approximation for multidimensional arrays

Cited in

(3)

• One-dimensional tensor network recovery
• An Eulerian approach to regularized JKO scheme with low-rank tensor decompositions for Bayesian inversion
• Solving the Fokker-Planck equation of discretized Dean-Kawasaki models with functional hierarchical tensor

This page was built for publication: High-dimensional density estimation with tensorizing flow

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6173686)