I started exploring the idea of using matrix eigenvalues as the "nonlinearity" in models, and wrote a second post in the series where I explore the scaling, robustness and interpretability properties of this kind of models. It's not surprising, but matrix spectral norms play a key role in robustness and interpretability.
I saw a lot of replies here for the previous post, so I hope you'll also enjoy the next post in this series:
https://alexshtf.github.io/2026/01/01/Spectrum-Props.html
reply