physical-reservoir-computing

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.