Sergey Petrov – Constructing effective tensor approximations using incomplete and noisy data
During the talk, wireless channel multidimensional tensor properties will be discussed: such a tensor, according to the multipath model, should have a decent low-rank canonical polyadic approximation. It will be shown that canonical approximations can be used to effectively solve the channel estimation and compression problems, as the canonical format is low-parametric and robust to measurement imperfections. In particular, analytic bounds will be provided for high-dimensional canonical approximation perturbations, and a channel extrapolation algorithm will be discussed that maintains acceptable quality for up to 20-160 milliseconds in advance.
Bio: Sergey Petrov has received a Master degree at Computational Mathematics and Cybernetics department of MSU, and continued as a PhD student at INM RAS. Currently – junior research scientist at INM RAS.