Network tuning by genetic algorithms
-
1
Norwegian University of Life Sciences, Mathematical Sciences and Technology, Norway
The dynamics and computational performance of neuronal networks depends critically on parameters characterizing the network architecture as well as the dynamics of synapses and single cells. For biologically realistic systems, the corresponding parameter space is high-dimensional. Anatomical and electrophysiological studies can confine the parameter space to biologically relevant regions only to some extent. The uncertainty in the choice of model parameters is still large. Fitting the model's dynamics to a desired target output is therefore a non-trivial task. Systematic scans through the full high-dimensional parameter space are not only computationally demanding and time consuming, but in many cases also not enlightening. Previous studies are therefore often restricted to low-dimensional sections through the parameter space or example parameter sets, thereby leaving the role of many parameters in the dark.
Optimization strategies like genetic algorithms provide a potential solution to this problem. They can help to reduce the degree of freedom in the choice of model parameters and to focus on biologically and functionally relevant subspaces. Here, we explore the applicability of genetic algorithms to the tuning of large networks of spiking neurons. As a benchmark system, we consider a random network of excitatory and inhibitory integrate-and-fire neurons with realistic connectivity (Brunel, 2000). Using a genetic algorithm, we search for optimal parameter values (synaptic weights and time constants, membrane time constants, in-degrees, spatiotemporal structure of external stimuli, etc.) resulting in network activity patterns with a defined target statistics (e.g. firing rate, inter-spike interval distribution, pairwise spike-train correlation). By studying the topography of the error landscape near the optimal solution, we investigate the sensitivity of the system to individual parameters (Gutenkunst et al., 2007) and to which extent parameters are redundant.
Acknowledgements:We acknowledge partial support by the Research Council of Norway (grant 178892 eNEURO). All network simulations were carried out with the neural simulation tool NEST (see http://www.nest-initiative.org).
References
1. Brunel N (2000) Dynamics of sparsely connected networks of excitatory and inhibitory spiking neurons. J Comput Neurosci 8(3): 183-208
2. Gutenkunst RN, et al. (2007) Universally Sloppy Parameter Sensitivities in Systems Biology Models. PLoS Comput Biol 3(10): e189
Conference:
Neuroinformatics 2009, Pilsen, Czechia, 6 Sep - 8 Sep, 2009.
Presentation Type:
Poster Presentation
Topic:
Computational neuroscience
Citation:
Enger
H,
Tetzlaff
T and
Einevoll
GT
(2019). Network tuning by genetic algorithms.
Front. Neuroinform.
Conference Abstract:
Neuroinformatics 2009.
doi: 10.3389/conf.neuro.11.2009.08.101
Copyright:
The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers.
They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.
The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.
Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.
For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.
Received:
25 May 2009;
Published Online:
09 May 2019.
*
Correspondence:
Hakon Enger, Norwegian University of Life Sciences, Mathematical Sciences and Technology, Akershus, Norway, hakon.enger@umb.no