Concentration of measure for non-linear random matrices with applications to neural networks and non-commutative polynomials

This provides theoretical tools for analyzing neural networks and non-commutative structures, but it appears incremental as it extends existing concentration results to new models.

The paper proves concentration inequalities for non-linear random matrices, with applications to estimating linear spectral statistics of neural network conjugate kernels and non-commutative polynomials in random matrices.