Modeling Spiking Neurons without Spikes

Published

August 2, 2024

Author(s)

Jeff Shainline, Bryce Primavera, Ryan O'Loughlin

Abstract

While spiking neuromorphic hardware holds promise for efficient implementations of artificial intelligence, the impact has been limited due in part to a lack of learning algorithms that achieve performance superior to conventional deep learning. One challenge related to developing algorithms for spiking neuromorphic hardware is that spikes are discontinuous events, which makes formal analysis difficult. Here we show that in the case of superconducting optoelectronic neurons, it is possible to construct dynamical equations that closely follow the full temporal response of the underlying spiking neurons without making any mathematical reference to the spikes themselves. The simple, leaky-integrator equations are continuous in time, opening possibilities for formal mathematical analysis and algorithm development without resorting to rate-coding approximations. Such a formalism is possible for superconducting neurons because signals are stored in the synapses and dendrites after the neuron fires, so a continuous value is available to enter the differential equations.

Proceedings Title

ACM ICONS 2024
International Conference on Neuromorphic Systems

Conference Dates

July 30-August 2, 2024

Conference Location

Arlington, VA, US

Pub Type

Conferences

Keywords

spiking neural network, neuromorphic computing, superconductors, photonics, superconducting optoelectronic networks

Citation

Shainline, J. , Primavera, B. and O'Loughlin, R. (2024), Modeling Spiking Neurons without Spikes, ACM ICONS 2024 International Conference on Neuromorphic Systems, Arlington, VA, US (Accessed February 23, 2026)