direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Inhalt des Dokuments

Modelle neuronaler Systeme

Ausgewählte Abstracts

Role of Colored Noise in the Detection of Synchronous Spiking Events in Neural Systems
Zitatschlüssel Hoch05
Autor Hoch, T. and Obermayer, K.
Buchtitel Computational and Systems Neuroscience
Jahr 2005
Zusammenfassung Recently, it has been shown that cortical neurons in the visual system are exposed to an enormous synaptic activity, which determines fundamental properties of the cell as for example the input resistance or the membrane time constant [1]. Furthermore they have found that the statistical properties of the membrane voltage depends critically on the actual synaptic input, which can change on a short time scale and thus leads to a strongly fluctuating membrane potential [2]. From modeling studies it is known that such a fluctuating membrane potential can improve the processing of sub-threshold signals, for example in a stochastic resonance setting [3,4]. In this contribution we ask the question, if a neuron or a population of neurons can react rapidly, reliably and temporally precise to transient, unexpected small inputs in such a noisy environment. Is it possible for a population to distinguish spikes initiated by small transient inputs from accidentally generated spikes? And is there a way for the neuron to optimize detection performance by altering the statistical properties of it's membrane voltage? The detection of transient changes in the input statistic can be considered as a basic neural computation and is important for coincidence detection as well as for the detection of synchronous spiking events in neural systems. Recently we have shown for a single leaky integrate-and-fire (LIF) neuron with current noise that if the noise is colored (Ornstein-Uhlenbeck process), pulse detection becomes more robust for increasing temporal correlations of the noise [5]. Given equal variance, the number of false positive events is reduced when temporal correlations increase, while the number of correctly detected pulses remains almost the same. To overcome the relatively low maximal detection rate of $P_cd = 0.5$ of a single neuron, a population of neurons has to be considered. This is also biologically realistic, since cortical processing is likely to be based on populations of cells rather than on single neurons [6]. Simulations with a population of LIF neurons showed that the detection performance (measured by the area under the ROC curve) improves rapidly with increasing population size and with increasing temporal correlation of the noise [7]. In this contribution we now present results from simulations with the biologically more realistic Hodgkin-Huxley (HH) point neurons. Similar to the case of the LIF model we find, that the rate of false positive events (the rate of "spontaneous" activity) decreases with increasing temporal correlations of the noise (balanced input and conductance noise, described by an Ornstein-Uhlenbeck process). In contrast to the LIF model, however, the probability of correct detection is no longer independent of the correlations of the noise. Because a signal with a stronger temporal correlation (larger value of the time constant in the Ornstein-Uhlenbeck process) is more smooth and has a lower average rate of change, the threshold of spike generation is increased (cf. Ref. [8]), leading to a reduction of the percentage of correctly detected pulses. This effect is a result of the gating mechanism of the active ion channels in the HH model. Depending on the model's parameters, a HH neuron can operate in two regimes. In one regime, the reduction of the spontaneous rate for increased temporal correlations dominates, and the smoothing of the noise leads to an improved overall pulse detection performance, similar to what is observed for the LIF model. In the other regime the reduction of the percentage of correctly detected pulses dominates, and pulse detection performance is best for short correlation times. In our contribution we will explore the complex relationship between detection performance, population size, conductance noise parameters, and the strength and shape of the input pulses. This work was supported by the DFG (SFB 618). [1] A. Destexhe, M. Rudolph and D. Paré, Nat. Rev. Neurosci. 4, 739-751, (2003). [2] A. Destexhe and D. Paré, J. Neurophysiol. 81, 1531-1547 (1999). [3] M. Rudolph and A. Destexhe, J. Comput. Neurosci. 11, 19-42 (2001). [4] D. F. Russel, L. A. Wilkens and F. Moss, Nature 402, 291-294 (1999). [5] G. Wenning, T. Hoch and K. Obermayer, Phys. Rev. E, in press [6] A. Pouget, R. S. Zemel and P. Dayan, Nat. Rev. Neurosci. 1, 125-132 (2000). [7] G. Wenning, T. Hoch and K. Obermayer, In Soc. Neurosci. Abstr. 30, CD-ROM (2004). [8] R. Azouz and C. M. Gray, PNAS 97, 8110-15 (2000).
Link zur Publikation Download Bibtex Eintrag

Zusatzinformationen / Extras

Direktzugang:

Schnellnavigation zur Seite über Nummerneingabe