edge-of-chaos

This paper presents a memory–stability–expressivity trilemma for trainable dissipative oscillator networks, showing that damping governs all three and limits trainability, with experimental validation on a 20-oscillator network confirming the theoretical bounds.

This paper develops a mean-field theory of dropout as a perturbation at the edge of chaos in neural networks, deriving scaling laws for correlation decay and establishing distinct universality classes for smooth and ReLU-like activations. It also yields optimal dropout scheduling that reduces test loss with no extra computational cost.