[LG] Random Matrix Theory for Deep Learning: Beyond Eigenvalues of Linear Models [Huazhong University of Science and Technolog & UC Berkeley] arxiv.org