Abstract

The discreet-time (map-based) approach to modeling nonlinear dynamics of spiking and spiking-bursting activity of neurons has demonstrated its very high efficiency in simulations of neuro-biologically realistic behavior both in large-scale network models for brain activity studies and in real-time operation of Central Pattern Generator network models for biomimetic robotics. This paper studies the next step in improving the model computational efficiency that includes quantization of model variables and makes the network models suitable for embedded solutions. We modify a map-based neuron model to enable simulations using only integer arithmetic and demonstrate a significant reduction of computation time in an embedded system using readily available, inexpensive ARM Cortex L4 microprocessors.

Highlights

  • The functionality of a brain and other neuronal systems of living organisms rely on the synaptic coupling organization among populations of neurons forming a function specific network of neurons with proper pathways of electrical activity in forms of short electrical spikes known as action potentials

  • They mainly rely on the development and implementation of a spiking neuron model and synapses that replicate the dynamics of neurons as close to real biological neurons as possible that fits in computational hardware

  • In the reported study we used neuronal and synaptic models scaled and coded using 32-bit integer variables and parameters. This increased the computational efficiency of the model Equation 1 and 2 about ten times and the overall network simulation, about five times

Read more

Summary

Introduction

The functionality of a brain and other neuronal systems of living organisms rely on the synaptic coupling organization among populations of neurons forming a function specific network of neurons with proper pathways of electrical activity in forms of short electrical spikes known as action potentials. Dynamical properties of spiking electrical activity of neurons, conditioned by the intrinsic properties of neurons or evoked in the response to external stimuli or other synaptic inputs, provide critical elements in shaping proper neuronal activity and, the functionality of neurobiological networks This spiking activity propagating along the synaptic pathways is a key element of information processing in the brain, classification of sensory inputs, decisionmaking and motor control (Shepherd, 2004). Numerous recent studies focus on the development of artificial (engineering) systems that use elements of spiking activity to create a network that could perform some elements of brain functionality, sensory processing, motor control (Dura-Bernal et al, 2014) and other functions They mainly rely on the development and implementation of a spiking neuron model and synapses that replicate the dynamics of neurons as close to real biological neurons as possible that fits in computational hardware. Two main directions of these studies are the use of analog computing simulating the network with specially designed VLSI integrated circuits (Indiveri et al, 2006; Silver et al, 2007) and numerical simulation of the model with specially designed microchips, GPUs FPGAs or DSPs (Yavuz et al, 2016; Zbrzeski et al, 2016; Cheung et al, 2016; Brette et al, 2007 and references therein)

Objectives
Conclusion
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call