Convergence of Two-Timescale RSGN Learning

Explore how timescale separation, Hebbian strength, and decay rate affect convergence of the coupled gradient descent + Hebbian structural learning system.

Epsilon (timescale ratio)

0.010

Hebbian Strength

0.30

Decay Rate

0.50

Init Radius

0.50

Converged

Contraction Rate

Topology Changes

Max Conformal Factor

Convergence of Two-Timescale RSGN Learning

Loss Trajectory (Fast Lyapunov)

Affinity Lyapunov (Slow)

Contraction Regime Map

Epsilon Sensitivity