abstract:
We would like to reconstruct
a tensor in the hierarchical tensor format,
(we take TT tensors as particular example for demonstration)
which is supposed to have given ranks $ (r_i ) $, from
$p$ linear measurements. Typically these measurements should be
values of the tensor at uniformly at random sampled points (tensor completion).
e would like show that under a RIP condition with
appropriately good RIP constants, an apropriate iterative hard thresholding algorithm converges the sought HT tensor. |