Fast Sampling for Linear Inverse Problems of Vectors and Tensors using Multilinear Extensions

This paper studies the problem of sampling vector and tensor signals, which is the process of choosing sites in vectors and tensors to place sensors for better recovery. A small core tensor and multiple factor matrices can be used to sparsely represent a dense higher-order tensor within a linear model. Using this linear model, one can effectively recover the whole signals from a limited number of measurements by solving linear inverse problems (LIPs). By providing the closed-form expressions of multilinear extensions for the frame potential of pruned matrices, we develop an algorithm named fast Frank-Wolfe algorithm (FFW) for sampling vectors and tensors with low complexity. We provide the approximation factor of our proposed algorithm for the factor matrices that are non-orthogonal and have elements of the same sign in each row. Moreover, we conduct experiments to verify the higher performance and lower complexity of our proposed algorithm for general factor matrix. Finally, we demonstrate that sampling by FFW and reconstruction by least squares methods yield better results for image data compared to convCNP completion with random sampling.