Parallel prefix scan for the computation of axonal projection patterns in biological neural networks

Sebastian James

doi:10.20935/al461

Abstract

Parallel prefix scan for the computation of axonal projection patterns in biological neural networks

Highlights

A complex neuron model in which the cell’s membrane voltage, its ion channels, and the release of various neurotransmitters may be modeled to understand the network’s behavior
The benefit which the graphical processing unit (GPU) can offer depends on the complexity of the neuron model and the connectivity of the network; N neurons in a network can be simulated in parallel by N processing threads for only one timestep before their outputs must be transferred according to the network’s connectivity
The GPU is wellsuited to the parallel execution of the neuron models, but not for the transfer of signals, which requires that GPU threads coordinate memory accesses

Summary

Introduction

A complex neuron model in which the cell’s membrane voltage, its ion channels, and the release of various neurotransmitters may be modeled to understand the network’s behavior. Researchers have often focussed on the compute time required for simulation of the network rather than for finding the connection parameters. Given the significant additional complexity of GPU code development, the adoption of GPU computation to simulate neurophysiological neural networks has been limited.

Results

Conclusion