Neural ODE
Continuous-depth network as a learned vector field — trajectory integration and phase portrait
Phase Portrait
Trajectories
Start
End
dz/dt = f_θ(z,t) = W₂·tanh(W₁·z + b₁) + b₂
Vector Field Controls
Preset:
Spiral Attractor
Vortex
Saddle Point
Limit Cycle
Double Well
Lorenz-like
W₁ scale
1.0
W₂ scale
1.0
Bias b₁
0.0
T (depth)
3.0
dt step
0.05
Animate
Redraw
Add Points
Loss Curve (spiral to center)