Regensburg 2019 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
BP: Fachverband Biologische Physik
BP 12: Poster II
BP 12.50: Poster
Dienstag, 2. April 2019, 14:00–16:00, Poster B2
Taming the bias when estimating correlations from spike recordings — Jens Wilting1, Johannes Zierenberg1, •Leonhard Leppin1,2, and Viola Priesemann1 — 1Max Planck Institute for Dynamics and Self-Organization, Göttingen, Germany — 2Georg-August-Universität Göttingen, Germany
What can we infer about a dynamical system if we can only observe a very small part of it? The problem of subsampling is common to the study of many systems. It is particularly severe in neuroscience, because electrophysiological recordings of spiking activity can only assess a small fraction of all neurons simultaneously. This subsampling has hindered characterizing even most basic properties of collective spiking in cortical networks. We proved that whenever a population is subsampled, the observed spike count cross-correlation between the populations can be strongly underestimated. The same holds for the autocorrelation strength of subsampled activity of a single population. These limitations hinder the correct inference of the underlying network dynamics. To overcome the systematic bias, we derived a novel estimator, which can infer properties of activity propagation even under strong subsampling. In this framework, the dynamical state is characterized by the average number of spikes triggered causally by a single spike in a neuron. Our generalization of the estimator to many populations now enables us to infer afferent contributions, recurrent propagation within a population, and reciprocal propagation between populations, and thereby enables us to contribute to identifying functional connections between brain areas.