The operations and maintenance of satellite networks heavily depend on traffic measurements.

This suggests that the low-rank fast attention only works for functions approx-imable by polynomials.

We prove the non-asymptotic convergence and inference guarantee for our method in general discrete distributions.

In order to address this limitation, we propose a novel class of functions that can characterize the loss landscape of modern deep models without requiring extensive over-parametrization and can also include saddle points.