% IMPORTANT: The following is UTF-8 encoded. This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.
@INPROCEEDINGS{Krishnan:281077,
author = {Krishnan, Jeyashree and Mana, PierGianLuca and Helias,
Moritz and Kunkel, Susanne and Di Napoli, Edoardo and
Diesmann, Markus},
title = {{D}etection of spiking events in continuous-time spiking
neuron models},
school = {RWTH Aachen},
reportid = {FZJ-2016-00782},
year = {2015},
abstract = {The leaky integrate-and-fire neuron model is one of the
commonly used spiking neuron models that can mimic the
dynamics of neurons to high accuracy. This model consists of
a system of first order linear differential equations with
which the sub-threshold dynamics can be exactly integrated.
Any excursion of the membrane potential above threshold
leads to a spike, immediately after which the membrane
potential is clamped to zero and input is sent to other
neurons.Theoretical descriptions of neuron models require
that the numerical implementations be in agreement with the
exact solutions of the mathematical model. In time-driven
simulators like NEST, the state of the neurons is updated in
discrete time steps, and the spikes are detected and emitted
only at the end of each step. Such a discrete-time neuronal
network simulation leads to the following problems:- Spikes
get constrained to the time grid, therefore they do not
carry a precise time stamp, leading to artifical
synchronization.- Depending on the computational step size,
it may happen that the neuron voltage is below threshold at
the beginning and end of the timestep with the excursion
happening within the step, thereby causing a spike miss. -
In addition, grid-constrained spiking causes an integration
error that decreases only linearly with the computational
step size.The purpose of this work is to formulate and
implement efficient techniques that can be included in the
neuron models to handle events at every point on the time
grid by computing the precise spike times. We have
constructed a precise numerical implementation of a
particular variant of the leaky integrate-and-fire neuron
model that does not miss any spikes. This implementation
relies on the computation of the exact time to the maximum
of potential in closed form based on the Lambert-W function.
This model can catch otherwise missed spikes for large
computational step sizes but is in principle computationally
expensive because the cost comes from both the frequency of
the test and the time for its calculation.To decrease the
computational cost of the spike test, we have constructed an
additional test that can predict the occurrence of threshold
crossing in continuous-time allowing us to calculate the
time to maximum from the previous test only when necessary.
This test is based on the analysis of the trajectories in
state-space governed by the system of equations describing
this model. This has helped us construct a series of tests
that can predict before propagation whether or not an event
is expected in the future, thereby helping us achieve both
high accuracy and less computational time.},
month = {Jan},
date = {2016-01-11},
organization = {3rd HBP Winter School, Manchester, UK,
Manchester (UK), 11 Jan 2016 - 15 Jan
2016},
subtyp = {Other},
cin = {INM-6 / JSC},
cid = {I:(DE-Juel1)INM-6-20090406 / I:(DE-Juel1)JSC-20090406},
pnm = {574 - Theory, modelling and simulation (POF3-574) /
Simulation and Data Laboratory Quantum Materials (SDLQM)
(SDLQM)},
pid = {G:(DE-HGF)POF3-574 / G:(DE-Juel1)SDLQM},
typ = {PUB:(DE-HGF)24},
url = {https://juser.fz-juelich.de/record/281077},
}