Oscillator-Based Processing Unit for Formant Recognition

Abstract

Oscillatory neural networks have been successfully applied to a number of computing problems, such as associative memories and computationally hard optimization tasks. In this paper, we show how to use oscillators to process time-dependent waveforms with minimal or no preprocessing. Since preprocessing and first-layer processing are often the most power-hungry steps in neural networks, our findings may open new doors to simple and power-efficient edge-AI devices.

19 pages, 13 figures

References in corpus (6)

• Vowel recognition with four coupled spin-torque nano-oscillators
• Wave Physics as an Analog Recurrent Neural Network
• Spontaneous synchronization of coupled oscillator systems with frequency adaptation
• Training Coupled Phase Oscillators as a Neuromorphic Platform using Equilibrium Propagation
• Reservoir Computing using High Order Synchronization of Coupled Oscillators
• An All-digital 8.6-nJ/Frame 65-nm Tsetlin Machine Image Classification Accelerator