Neural ODEs on Biological Dynamics
Neural ODEs applied to the FitzHugh-Nagumo neuron model — phase-portrait recovery as a more honest diagnostic than trajectory MSE.
Part 1
FitzHugh-Nagumo & Why Continuity Matters
The FHN equations, autonomous vs non-autonomous dynamics, why continuous-time inductive biases match biological data.
Part 2
Implementation
Ground-truth FHN simulator, autonomous Neural ODE with RK4, discrete-RNN baseline.
View Code on GitHub
Part 3
Phase-Portrait Recovery
Both models fit trajectories to MSE 0.02; only the Neural ODE recovers the underlying flow field (cosine 0.94–0.96) — that's the difference between fitting data and learning the physics.