Quadratic Suffices for Over-parametrization via Matrix Chernoff Bound

This work addresses theoretical guarantees for over-parametrization in deep learning, which is incremental as it builds on existing results.

The paper tackles the problem of over-parametrization size in deep learning theory, improving upon prior results by Li and Liang (2018) and Du et al. (2019), with the outcome being a reduction to quadratic sufficiency as indicated by the title.