Leonidas Elefteriou, BCCN Berlin

Information filtering in stochastic neuron models with slow intrinsic noise

  For all living beings, processing of external stimuli is of vital importance. This can be realized on two stages: firstly on the level of single neurons and secondly on the level of neuronal networks. Processing of time dependent external stimuli can lead to preferences regarding i) slow, ii) intermediate or iii) fast signal components. A proper framework to analyze the information processing/filtering characteristics is Shannon’s information theory. Based on the spectral coherence function, one can classify three filtering characteristics: i) low-pass, ii) band-pass, and iii) high-pass filter on information.
   First, the information filtering characteristics of the conductance based Traub-Miles model equipped with slow and stochastic adaptation channels is investigated numerically. These adaptation channels give rise to positive correlations between interspike intervals (ISIs), which can lead to band-pass filtering on information. This effect is studied in different firing regimes. Second, a simplified and analytically tractable threshold neuron model is studied, which mimics essential characteristics of the Traub-Miles model: i) ISI distribution, ii) exponentially correlated ISIs. In this neuron model, positive ISI correlations are implemented by employing a stochastic threshold that changes according to an autoregressive processes. It is confirmed numerically that positive ISI correlations can lead to band-pass filtering on information in this model as well. In order to distinguish between low and band/high pass filter, the curvature of the coherence function at zero frequency is used and analytical expressions are derived, which involve analytical approximations of all necessary spectra, especially the power spectrum of the spike train. Finally, both models (Traub-Miles and threshold model) are compared and the results are summarized. Additionally, an outlook to future work in order to match the two models even further is presented and additional mechanisms beside positive ISI correlations are briefly presented.

Additional Information

Master thesis defence in the Master Computational Neuroscience

Organized by

Benjamin Lindner / Robert Martin