International Conference on the spectral theory of the tensor
Hongqin Li,Qingzhi Yang,Shmuel Friedland,Joshua Cooper,Wen Li,Yaotang Li,Yuhong Dai,Liping Zhang,Xiaohuan Mo,Zhening Li,Yinyu Ye,Zhi-Quan Luo,Dustin Cartwright,Wenyu Sun,Naihua Xiu,Simai He,Xinzhen Zhang
2012-01-01
Abstract:A tensor singular value decomposition based upon the *-product defined on third-order tensors by Kilmer, Martin and Perrone provides an effective means of generalizing Principal Component Analysis. This talk will focus on tensor eigendecompostions and generalized eigendecompostions under the *-product. Discusion will include an overview of the theoretical underpinings, some computational concerns and also the use of tensor eigendecompositions in a generalization of Fischer's Linear Discriminant Analysis as applied to object recognition. The number of eigenvalues of a tensor Dustin Cartwright (Yale University / MPI) Abstract I will discuss the computation of the number of eigenpairs of a generic tensor. I will also discuss what may happen for special tensors, for which the number of eigenpairs may coincide or their number may become infinite. These questions are related to the structure of the characteristic polynomial of a tensor.I will discuss the computation of the number of eigenpairs of a generic tensor. I will also discuss what may happen for special tensors, for which the number of eigenpairs may coincide or their number may become infinite. These questions are related to the structure of the characteristic polynomial of a tensor. On the Z-eigenvalues of the signless Laplacian tensor for an even uniform hypergraph An Chang (Fuzhou University) Abstract In last two decades, the spectral graph theory has become one of the most active branches in the graph theory duo to its various applications in many disciplines. In fact, it is the branch of mathematics that studies graphs by using algebraic properties of associated matrices, such as the adjacency matrix or Laplace matrix. As graphs are related to matrices , hypergraphs are in a natural way related to tensors which could reveal more higher order structures than matrices. In this talk, we generalize the signless Laplacian matrices for graphs to the signless Laplacian tensors for even uniform hypergraphs and set some fundamental properties for the spectral hypergraph theory based upon the signless Laplacian tensors. Especially, the smallest and the largest Z-eigenvalues of the signless Laplacian tensor for an even uniform hypergraph are studied and their connections with hypergraph bipartition, maximum degree and edge cut are discussed.In last two decades, the spectral graph theory has become one of the most active branches in the graph theory duo to its various applications in many disciplines. In fact, it is the branch of mathematics that studies graphs by using algebraic properties of associated matrices, such as the adjacency matrix or Laplace matrix. As graphs are related to matrices , hypergraphs are in a natural way related to tensors which could reveal more higher order structures than matrices. In this talk, we generalize the signless Laplacian matrices for graphs to the signless Laplacian tensors for even uniform hypergraphs and set some fundamental properties for the spectral hypergraph theory based upon the signless Laplacian tensors. Especially, the smallest and the largest Z-eigenvalues of the signless Laplacian tensor for an even uniform hypergraph are studied and their connections with hypergraph bipartition, maximum degree and edge cut are discussed. Eigenvalues for stochastic tensors Kungching Chang (Peking University) Abstract Basic properties of transition probability tensors, introduced by Ng et al., are studied. The uniqueness of the positive eigenvector is not necessarily true in general. Sufficient conditions on this class of tensors are investigated to ensure the uniqueness of positive Z-eigenvector. Three different methods: Contraction mappings, monotone operator, and the Brouwer index of fixed points, are applied in this study. This is a joint work with T. Zhang.Basic properties of transition probability tensors, introduced by Ng et al., are studied. The uniqueness of the positive eigenvector is not necessarily true in general. Sufficient conditions on this class of tensors are investigated to ensure the uniqueness of positive Z-eigenvector. Three different methods: Contraction mappings, monotone operator, and the Brouwer index of fixed points, are applied in this study. This is a joint work with T. Zhang. Spectra of Hypergraphs Joshua Cooper (University of South Carolina) Abstract We present a spectral theory of hypergraphs that closely parallels graph spectral theory. Classic work by Gel'fand-Kapranov-Zelevinsky and Canny, as well as more recent developments by Chang, Lim, Pearson, Qi, Zhang, and others has led to a rich understanding of “hyperdeterminants” of hypermatrices, a.k.a. multidimensional arrays. Hyperdeterminants share many properties with determinants, but the context of multilinear algebra is substantially more complicated than the linear algebra required to understand spectral graph theory (i.e., ordinary matrices). Nonetheless, it is possible to define eigenvalues of a tensor via its characteristic polynomial and variationally. We apply this notion to the “adjacency hypermatrix” of a uniform hypergraph, and prove a number of natural analogues of graph theoretic results. Computations are particularly cumbersome with hyperdeterminants, so we discuss software developed in Sage which can perform basic calculations on small hypergraphs. Open problems abound, and we present a few directions for further research. Joint work with Aaron Dutle of the University of South Carolina.We present a spectral theory of hypergraphs that closely parallels graph spectral theory. Classic work by Gel'fand-Kapranov-Zelevinsky and Canny, as well as more recent developments by Chang, Lim, Pearson, Qi, Zhang, and others has led to a rich understanding of “hyperdeterminants” of hypermatrices, a.k.a. multidimensional arrays. Hyperdeterminants share many properties with determinants, but the context of multilinear algebra is substantially more complicated than the linear algebra required to understand spectral graph theory (i.e., ordinary matrices). Nonetheless, it is possible to define eigenvalues of a tensor via its characteristic polynomial and variationally. We apply this notion to the “adjacency hypermatrix” of a uniform hypergraph, and prove a number of natural analogues of graph theoretic results. Computations are particularly cumbersome with hyperdeterminants, so we discuss software developed in Sage which can perform basic calculations on small hypergraphs. Open problems abound, and we present a few directions for further research. Joint work with Aaron Dutle of the University of South Carolina. A subspace projection method for finding the extreme Z-eigenvalues of supersymmetric positive definite tensor Yu-Hong Dai (AMSS,Chinese Academy of Sciences) Abstract A subspace projection method is presented for finding the exteme Z-eigenvalues of m-th order n-dimensional supersymmetric positive definite tensor $A$. The idea is based on the geometric properties of Z-eigenvalues and Z-eigenvectors that the shortest distance from $S=\{x\inR^n|Ax^m=c,c>0\}$ to the origin is $\sigma_{min}=(\frac{c}{\lambda_{max}})^{\frac{1}{m}}$. And the shortest distance can be obtained by projecting the original problem into a two dimension subspace in each iteration where the Z-eigenvalue of the reduced tensor can be computed directly. A subspace projection algorithm is proposed. The preliminary numerical results show that the algorithm performs very well. Some extensions are also considered. This is a joint work with Chunlin Hao.A subspace projection method is presented for finding the exteme Z-eigenvalues of m-th order n-dimensional supersymmetric positive definite tensor $A$. The idea is based on the geometric properties of Z-eigenvalues and Z-eigenvectors that the shortest distance from $S=\{x\inR^n|Ax^m=c,c>0\}$ to the origin is $\sigma_{min}=(\frac{c}{\lambda_{max}})^{\frac{1}{m}}$. And the shortest distance can be obtained by projecting the original problem into a two dimension subspace in each iteration where the Z-eigenvalue of the reduced tensor can be computed directly. A subspace projection algorithm is proposed. The preliminary numerical results show that the algorithm performs very well. Some extensions are also considered. This is a joint work with Chunlin Hao. Canonical Polyadic Decomposition and Block Term Decompositions: Uniqueness and Signal Separation Lieven De Lathauwer (KU Leuven) Abstract Canonical Polyadic Decomposition (CPD) writes a higher-order tensor as a minimal sum of rank-1 tensors. Kruskal derived the following uniqueness result. Let A, B, C denote the factor matrices in a PD of a tensor T, and let kA + kB + kC ≥ 2R+2, where kA, kB, kC denote the k-rank of A, B, C, respectively, then R is the rank of T, i.e., the PD is canonical, and this decomposition is unique. Other uniqueness conditions have been derived by Jiang and Sidiropoulos, and by De Lathauwer. These conditions are on one hand more restrictive than Kruskal’s in the sense that one of the factor matrices, say C, needs to have full column rank, which implies that the rank R may not exceed all tensor dimensions. On the other hand, the results are an order of magnitude more relaxed than Kruskal’s in terms of the conditions on A and B. In this talk we present new CPD uniqueness conditions, connecting Jiang – Sidiropoulos – De Lathauwer’s results with Kruskal’s result. We relax the condition on C, no longer requiring that it has full column rank, while making the conditions on A and B more restrictive. Part of the study leads to new Kruskal-type conditions that are more relaxed than the original. For instance, we explain that, if kA + rB + rC ≥ 2R+2, and rA + kB + rC ≥ 2R+2, and rA + rB + kC ≥ 2R+2, where rA, rB, rC denote the rank of A, B, C, respectively, then the PD is canonical and the decomposition is unique. Recently introduced Blok Term Decompositions (BTD) write a higher-order tensor as a minimal sum of tensors of certain low multilinear rank. We present generalization