Model-free Inference of Interaction Circuits from Nonlinear Dynamics?

Date

Aug 18, 2016

Time

2:00 PM - 3:00 PM

Speaker

Marc Timme (Network Dynamics

Affiliation

TU Darmstadt and MPI for Dynamics and Self-Organization )

Language

en

Main Topic

Physik

Host

Advanced Study Group

Description

Revealing direct physical interactions among the units of a network constitutes a basic prerequisite for understanding fundamental features of its structure and function. As faithful models for many natural networks are missing, identifying such interactions is often not possible without influencing the systems' dynamics. Here we present a framework for revealing physical network interactions based on observed time series. Representing pairwise interactions by an explicit dependency matrix enables a systematic expansion of the dynamical rules in basis functions and thereby confines potential network structures by a linear system of equations. A greedy numerical method finds solutions that are block-sparse with respect to each pair of units. We generalize the concept to a neural circuit setting where instead of the time-continuous state variables, only pulse-sequences at discrete times, and thus small parts of the dynamics' can be observed. These results enable us to reveal interaction networks for systems where a detailed model is not known and only indicators of state variables are observable. Introductory Review: Timme & Casadiego, J. Phys. A (2014) http://dx.doi.org/10.1088/1751-8113/47/34/343001 This is work with Jose Casadiego and Dimitra Maoutsa.

Last modified: Aug 18, 2016, 9:44:10 AM