- Dynamical Isometry and a Mean Field Theory of CNNs: How to Train 10,000-Layer Vanilla Convolutional Neural Networks
- A Geometric Analysis of Neural Collapse with Unconstrained Features
- The emergence of spectral universality in deep networks
- NIPS 2017 Spectral Representations for Convolutional Neural Networks
- https://dl.acm.org/doi/pdf/10.5555/3546258.3546530, 2.4
- [2104.13478] Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges