Concentration of measure for nonlinear random matrices
- Speaker(s)
- Radosław Adamczak
- Affiliation
- University of Warsaw
- Language of the talk
- English
- Date
- Oct. 16, 2025, 12:15 p.m.
- Room
- room 3160
- Title in Polish
- Concentration of measure for nonlinear random matrices
- Seminar
- Seminar of Probability Group
I will present concentration inequalities for certain models of random matrices related to neural networks. I will discuss their optimality and dependence on the choice of activation functions and the distribution of weights. Next, I will apply them to eigenvalue statistics, in particular showing a deviation inequality for the first Wasserstein distance between the empirical and expected spectral measures of the conjugate kernel of the neural network.
You are not logged in |