Stable neural networks and connections to continuous dynamical systems
Matthias J. Ehrhardt, Davide Murari, Ferdia Sherry · 2025
Read Stable neural networks and connections to continuous dynamical systems by Matthias J. Ehrhardt, Davide Murari, Ferdia Sherry (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-22299.txt.
Description
Matthias J. Ehrhardt, Davide Murari, and Ferdia Sherry survey a design philosophy for neural networks. It treats a network as a dynamical system unfolding in time. The motivation is stability. Networks that a tiny crafted input change can fool are untrustworthy. One active school of defense borrows the mathematics of continuous dynamics and optimal control.
The paper gives a bird's-eye map of that school. The fundamental concepts most of the literature shares are named. It then slows down for one concrete approach to stable networks. The theory is laid out, and the implementation is described. Working code accompanies the text. A notebook shows a small adversarial robustness example for image classification. It runs on an ordinary computer. The discussion walks through the toy example so readers can feel the stability mechanism.
Catalog record
| CREATOR | Matthias J. Ehrhardt, Davide Murari, Ferdia Sherry |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 80 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-22299