【摘 要】
：
Low rank tensor learning, such as low rank tensor approximation, tensor completion, as well as multilinear multitask learning, draws much attention in recen
【出 处】
：
2016年张量和矩阵学术研讨会(International conference on Tensor, Matrix a
论文部分内容阅读
Low rank tensor learning, such as low rank tensor approximation, tensor completion, as well as multilinear multitask learning, draws much attention in recent years. In this talk, we introduce algorithms, based on rank-one tensor updating, for solving the problems. At each iteration of the proposed algorithms, the main cost is only to compute a rank-one tensor, by approximately solving a tensor spectral norm problem. Comparing with matrix SVD based state-of-the-art methods, computing the rank-one tensor can be much cheaper, resulting into the efficiency of the proposed methods. Linear convergence rate is then established, either with a convex or nonconvex cost function. Experiments on (robust) tensor completion and multilinear multitask learning demonstrate the efficiency and effectiveness of the proposed algorithms.
其他文献
Identifying the positive definiteness of an even-order homogeneous multivariate form is an important task due to its wide applications in such as medical im
In this paper, some new Brauer-type eigenvalue inclusion theorems are established for general tensors. We show that new eigenvalue inclusion sets are sharpe
In this talk, we first review some results related Sylvester-type matrix equations and matrix decompositions. Then we investigate and analyze in detail the
A Hankel tensor is called a strong Hankel tensor if the Hankel matrix generated by its generating vector is positive semi-definite. It is known that an even
An iterative method for calculating the largest eigenvalue of a nonnegative tensor was proposed by Ng, Qi and Zhou in 2009. In this paper, two extrapolated
In this presentation, we would like to investigate some adaptive gradient methods for generalized tensor eigenvalue complementarity problem (TEiCP). Firstly
Computing a few eigenpairs from large-scale symmetric eigenvalue problems is far beyond the tractability of classic eigensolvers when the storage of the eig
This talk is concerned with solving some structured multi-linear systems, especially focusing on the equations whose coeffcient tensors are M-tensors, or ca
In this talk I will talk about the latest development of applications of tensor theory to probability and statistics. Those include the moment tensor of ran
Transition probability tensors of order 3 in dimension 3 and of order 4 in dimension 2 are studied. In both cases, we prove that an irreducible symmetric tr