Rechnen mit Oszillationen und Wellen: Von der Dynamik der Großhirnrinde zu analoger KI

We perform simulations of recurrent networks inspired by the mammalian cerebral cortex and discovered that our harmonic oscillator recurrent network (HORN) model captures essential properties of its dynamics, revealing a novel computational strategy that exploits synchrony, resonance, phase shifts and interference of traveling waves for fast, highly parallel computations. We implemented the HORN principle in analog electronic hardware, showing that it is a powerful computational substrate while at the same time being robust and highly energy efficient.