This talk describes an algorithm for computing a low-Tucker-rank approximation of a tensor.
The method applies a randomized linear map to the tensor to obtain a sketch
that captures the important directions within each mode, as well as the interactions among the modes.
The sketch can be extracted from streaming or distributed data or with a single pass over the tensor,
and it uses storage proportional to the degrees of freedom in the output Tucker approximation.
The algorithm does not require a second pass over the tensor, although it can exploit another view
to compute a superior approximation. The paper provides a rigorous theoretical guarantee
on the approximation error. Numerical experiments show that that the algorithm
produces useful results that improve on the state-of-the-art for streaming Tucker decomposition.
Copyright. All Rights Reserved.