Finite Time Lyapunov Exponents for Neural ODEs

This repository contains the code for the paper:
Tracking Finite-Time Lyapunov Exponents to Robustify Neural ODEs

The code implements finite time Lyapunov exponents (FTLE) for low-dimensional neural ODEs. Various Jupyter notebooks track Lyapunov exponents for different nODE dynamics and demonstrate a modified training method (FTLE suppression) that improves robustness.

Getting Started: MLE_master.ipynb is the best starting point.

---

Method Comparison: Robustness via FTLE Suppression

Standard Training	Training with FTLE suppression

---

FTLE Dynamics Visualized

Lyapunov exponents for non-autonomous nODE


Lyapunov exponents for autonomous nODE

---

Built using the torchdiffeq package.