Skip to content

rhshi/TensorDecomposition.jl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

26 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

TensorDecomposition

Build Status

This repository is Julia code for the paper: Bobby Shi, Julia Lindberg, Joe Kileel, "Efficient Tensor Decomposition via Moment Matrix Extension", arXiv:2506.22564 (2025). This code includes an implementation of efficient moment matrix extension via linear algebraic operations, requiring no symbolic algebra packages. Also included is an implementation of monomial decomposition via a paramaterization of the space of decompositions.

Proofs

As described in the paper, included are computer assisted proofs for efficient formats in proofs/4-17.jld2 for $n=4, \dots, 17$; for each $n$ we give a prime $p$ and an explicit specialization of points. We use the computer algebra package Nemo to perform computations over finite fields; some examples are given in proofs/example.ipynb, giving specializations such that $\mathbf{A}$ is full column rank over finite fields.

Examples

All examples are included in the folder examples:

  • d_4.ipynb: An example for $d=4$. Includes an example of efficient decomposition for $n=10, r=51$ where the matrix $\mathbf{A}$ is generically full column rank. Also includes an example where a decomposition has three colinear points so the decomposition is not unique; decompositions are given via a parameterization of the space of decompositions.
  • d_6.ipynb: An example for $d=6$. Includes an example of efficient decomposition for $n=5, r=50$ where the matrix $\mathbf{A}$ is generically full column rank. Also includes an example of a tensor where a set $B$ cannot be chosen to be the first $r$ monomials yet the tensor is still efficiently decomposable with a unique decomposition.
  • monomial.ipynb: An example of monomial decomposition, giving multiple decompositions via choosing different sets of parameters.
  • binary.ipynb: An example of binary tensor decomposition, specifically Example 3.20 from the paper.
  • linearly_dependent.ipynb: An example of a tensor format ($n=5, r=16$) such that the matrix $\mathbf{A}$ is tall but is never full column rank numerically.

Packages

Julia is required with the following packages:

  • Combinatorics
  • Graphs
  • LinearAlgebra
  • Random
  • LinearSolve
  • SparseArrays
  • Nemo
  • JLD2
  • IJulia

About

Supplementary code for the paper "Efficient Tensor Decomposition via Moment Matrix Extension"

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors