Volume 9 Supplement 1

Seventeenth Annual Computational Neuroscience Meeting: CNS*2008

Open Access

Analysis of stochastic integration with a network of bistable units

BMC Neuroscience20089(Suppl 1):P16

DOI: 10.1186/1471-2202-9-S1-P16

Published: 11 July 2008

The ability to integrate over time and to keep a memory of the result of this computation is thought to be important for several operations performed by the nervous system. Several models have been proposed for the biological implementation of the integration of inputs. Some of these models consist of a set of units which have two stable states in the absence of noise: an activated state, characterized by high activity and a spontaneous state, characterized by low activity. In these models the value of the integral is approximated by the number of bistable units that will reach and stay in the activated state given a certain input. Using such a network both deterministic [1] and stochastic mechanisms [2] have been proposed for integration of inputs. Further, experimental observations were found to be consistent with this type of models [2]. In this work we analyze theoretically how a network of stochastic bistable units performs integration of an external input. We use a simple probabilistic model to study how the number of activated units evolves with time, given a certain input. We consider both unconnected and connected units. For the unconnected model we derive analytical expressions for the ability of a system to discriminate between inputs with different durations or strengths. The difference between two inputs needed to assure that they can be discriminated depends on the absolute strengths of the inputs. We show that the connected system can, on average, perform an almost perfect integration, but that the performance on single trials is strongly dependent on the specifications of the model and inputs. Finally, we relate the probabilistic model with more biologically plausible systems, where the bistable units are neuronal populations or neurons which can be described by their average firing rates or by their spiking activities.

Declarations

Acknowledgements

R.A. acknowledges the financial support of the European research project DiM (FP6-IST, Contract No. 027198). A. L. acknowledges the financial support of the Spanish Ministry of Education and Science (Juan de la Cierva and Ramon y Cajal programs).

Authors’ Affiliations

(1)
Computational Neuroscience Group, Universitat Pompeu Fabra

References

  1. Koulakov AA, Raghavachari S, Kepecs A, Lisman JE: Model for a robust neural integrator. Nature Neurosci. 2002, 5 (8): 775-782. 10.1038/nn893.View ArticlePubMedGoogle Scholar
  2. Okamoto H, Isomura Y, Takada M, Fukai T: Temporal integration by stochastic recurrent network dynamics with bimodal neurons. J Neurophysiol. 2007, 97: 3859-3867. 10.1152/jn.01100.2006.View ArticlePubMedGoogle Scholar

Copyright

© Almeida and Ledberg; licensee BioMed Central Ltd. 2008

This article is published under license to BioMed Central Ltd.

Advertisement