Accelerated simulation of a neuronal population via mathematical model order reduction
Tutkimustuotos › › vertaisarvioitu
Yksityiskohdat
| Alkuperäiskieli | Englanti |
|---|---|
| Otsikko | Proceedings of the IEEE International Conference on Artificial Intelligence Circuits and Systems |
| Kustantaja | IEEE Xplore |
| Tila | Hyväksytty/In press - 10 joulukuuta 2019 |
| OKM-julkaisutyyppi | A4 Artikkeli konferenssijulkaisussa |
| Tapahtuma | IEEE International Conference on Artificial Intelligence Circuits and Systems - Kesto: 9 tammikuuta 2020 → … |
Conference
| Conference | IEEE International Conference on Artificial Intelligence Circuits and Systems |
|---|---|
| Ajanjakso | 9/01/20 → … |
Tiivistelmä
Mathematical modeling of biological neuronal net-
works is important in order to increase understanding of the
brain and develop systems capable of brain-like learning. While
mathematical analysis of these comprehensive, stochastic, and
complex models is intractable, and their numerical simulation
is very resource intensive, mean-field modeling is an effective
tool in enabling the analysis of these models. The mean-field
approach allows the study of populations of biophysically detailed
neurons with some assumptions of the mean behaviour of the
population, but ultimately requires numerical solving of high-
dimensional differential equation systems. Mathematical model
order reduction methods can be employed to accelerate the anal-
ysis of high-dimensional nonlinear models with a purely software-
based approach. Here we compare state-of-the-art methods for
improving the simulation time of a neuronal mean-field model
and show that a nonlinear Fokker-Planck-McKean-Vlasov model
can be accurately approximated in low-dimensional subspaces
with these methods. Using Proper Orthogonal Decomposition
and different variations of the Discrete Empirical Interpolation
Method, we improved the simulation time by over three orders
of magnitude while achieving low approximation error.
works is important in order to increase understanding of the
brain and develop systems capable of brain-like learning. While
mathematical analysis of these comprehensive, stochastic, and
complex models is intractable, and their numerical simulation
is very resource intensive, mean-field modeling is an effective
tool in enabling the analysis of these models. The mean-field
approach allows the study of populations of biophysically detailed
neurons with some assumptions of the mean behaviour of the
population, but ultimately requires numerical solving of high-
dimensional differential equation systems. Mathematical model
order reduction methods can be employed to accelerate the anal-
ysis of high-dimensional nonlinear models with a purely software-
based approach. Here we compare state-of-the-art methods for
improving the simulation time of a neuronal mean-field model
and show that a nonlinear Fokker-Planck-McKean-Vlasov model
can be accurately approximated in low-dimensional subspaces
with these methods. Using Proper Orthogonal Decomposition
and different variations of the Discrete Empirical Interpolation
Method, we improved the simulation time by over three orders
of magnitude while achieving low approximation error.