OSCILLATORY DYNAMICS AS A UNIVERSAL SUBSTRATE FOR COMPUTATION: FROM NEURAL CIRCUITS TO ARTIFICIAL INTELLIGENCE

While oscillatory dynamics are ubiquitous in biological neural networks, their precise computational role has remained a subject of intense debate. Traditionally viewed as potentially epiphenomenal, recent evidence from modelling studies suggests that oscillations constitute a fundamental mechanism for information processing in neural systems. This research synthesizes insights from theoretical, computational, and physical domains to establish oscillatory transients as a powerful substrate for computation across neural, artificial, and physical systems. We introduce the Harmonic Oscillator Recurrent Network (HORN) model, which models cortical microcircuits as coupled damped harmonic oscillators (DHO). Our simulations demonstrate that HORNs provide significant advantages over conventional, non-oscillatory recurrent architectures, including enhanced learning speed, parameter efficiency, and robustness. These benefits stem from unique computational principles: dual coding through amplitude and phase, resonance-based feature selectivity, and the ability to maintain fading memory for temporal integration using transient dynamics. Furthermore, we demonstrate the feasibility of these principles through an analog-electronic implementation, achieving high energy efficiency by utilizing wave interference for computation. This framework extends beyond neural substrates, finding parallels in photonic, mechanical, and fluid systems. Our findings suggest that wave-based dynamics provide a unifying substrate for analog computation, bridging the gap between biological intelligence and next-generation artificial intelligence systems.