Tensor principal component analysis via convex optimization
From MaRDI portal
Publication:2340337
DOI10.1007/S10107-014-0774-0zbMATH Open1312.65008arXiv1212.2702OpenAlexW2069287942MaRDI QIDQ2340337FDOQ2340337
Authors: Bo Jiang, Shiqian Ma, Shuzhong Zhang 
Publication date: 16 April 2015
Published in: Mathematical Programming. Series A. Series B (Search for Journal in Brave)
Abstract: This paper is concerned with the computation of the principal components for a general tensor, known as the tensor principal component analysis (PCA) problem. We show that the general tensor PCA problem is reducible to its special case where the tensor in question is super-symmetric with an even degree. In that case, the tensor can be embedded into a symmetric matrix. We prove that if the tensor is rank-one, then the embedded matrix must be rank-one too, and vice versa. The tensor PCA problem can thus be solved by means of matrix optimization under a rank-one constraint, for which we propose two solution methods: (1) imposing a nuclear norm penalty in the objective to enforce a low-rank solution; (2) relaxing the rank-one constraint by Semidefinite Programming. Interestingly, our experiments show that both methods yield a rank-one solution with high probability, thereby solving the original tensor PCA problem to optimality with high probability. To further cope with the size of the resulting convex optimization models, we propose to use the alternating direction method of multipliers, which reduces significantly the computational efforts. Various extensions of the model are considered as well.
Full work available at URL: https://arxiv.org/abs/1212.2702
Recommendations
- Tensor completion and low-\(n\)-rank tensor recovery via convex optimization
- Rank-1 tensor properties with applications to a class of tensor optimization problems
- Low-rank approximation and completion of positive tensors
- On tensor completion via nuclear norm minimization
- Numerical optimization for symmetric tensor decomposition
tensorprincipal component analysisalternating direction method of multipliersnuclear normnumerical exampleslow ranksemidefinite programming relaxation
Factor analysis and principal components; correspondence analysis (62H25) Multilinear algebra, tensor calculus (15A69) Semidefinite programming (90C22)
Cites Work
Cited In (36)
- Title not available (Why is that?)
- Managing randomization in the multi-block alternating direction method of multipliers for quadratic optimization
- Parallel matrix factorization for low-rank tensor completion
- Approximation algorithms for optimization of real-valued general conjugate complex forms
- Approximation methods for complex polynomial optimization
- The low-rank approximation of fourth-order partial-symmetric and conjugate partial-symmetric tensor
- Computing the largest C-eigenvalue of a tensor using convex relaxation
- The nonconvex tensor robust principal component analysis approximation model via the weighted \(\ell_p\)-norm regularization
- Successive Rank-One Approximations for Nearly Orthogonally Decomposable Symmetric Tensors
- On the optimization landscape of tensor decompositions
- Tensor Networks for Dimensionality Reduction and Large-scale Optimization: Part 2 Applications and Future Perspectives
- A hybrid second-order method for homogenous polynomial optimization over unit sphere
- On decompositions and approximations of conjugate partial-symmetric tensors
- Optimization on the hierarchical Tucker manifold - applications to tensor completion
- A DCA-Newton method for quartic minimization over the sphere
- Tensor Methods for Equality Constrained Optimization
- Optimization landscape of Tucker decomposition
- Approximating Tensor Norms via Sphere Covering: Bridging the Gap between Primal and Dual
- A Lagrange–Newton algorithm for tensor sparse principal component analysis
- Tensor Dictionary Learning for Positive Definite Matrices
- Characterizing real-valued multivariate complex polynomials and their symmetric tensor representations
- An approximation method of CP rank for third-order tensor completion
- The generalized degrees of freedom of multilinear principal component analysis
- On approximation algorithm for orthogonal low-rank tensor approximation
- Several approximation algorithms for sparse best rank-1 approximation to higher-order tensors
- Tensor symmetrization and its applications in generalized principal component analysis
- Parallel active subspace decomposition for tensor robust principal component analysis
- Tensor Canonical Correlation Analysis With Convergence and Statistical Guarantees
- Shifted eigenvalue decomposition method for computing C-eigenvalues of a piezoelectric-type tensor
- A note on semidefinite programming relaxations for polynomial optimization over a single sphere
- On norm compression inequalities for partitioned block tensors
- Title not available (Why is that?)
- Properties and methods for finding the best rank-one approximation to higher-order tensors
- Title not available (Why is that?)
- Rank-1 tensor properties with applications to a class of tensor optimization problems
- On cones of nonnegative quartic forms
Uses Software
This page was built for publication: Tensor principal component analysis via convex optimization
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2340337)
