Max Planck Institute for Dynamics and Self-Organization -- Department for Nonlinear Dynamics and Network Dynamics Group
Personal tools
Log in

BCCN AG-Seminar

Tuesday, 09.12.2014 17 c.t.

Encoding and discrimination of multi-dimensional signals using chaotic spiking activity

by Dr. Guillaume Lajoie
from Max Planck Insitute for Dynamics and Self-Organization

Contact person: Fred Wolf


Ludwig Prandtl lecture hall


Networks of neurons throughout the brain show highly recurrent connectivity. While this recurrence presumably contributes to information coding and processing, it also has a seemingly contrary effect: across many models, recurrent connectivity leads to chaotic dynamics. This implies that the spike times elicited by a given stimulus can and do depend sensitively on initial conditions or system perturbations, and suggests that precise temporal patterns of spikes are too fragile and unreliable to carry any useful information. However, recent work has shown that despite being chaotic, the variability of network responses to temporal inputs can be surprisingly low, a fact attributed to low-dimensional chaotic attractors and compatible with experimental observations of spike-time repeatability in recurrent networks. However, it is still unclear such complex dynamics may be used to encode specific features of inputs signals. In this talk, I will discuss the use of Random Dynamical Systems Theory as a framework to study information processing in high-dimensional, non-autonomous systems. Using a neural network model, I will present an overview of recent results linking random strange attractors to noise entropy and input discrimination of dynamical observables. Specifically, we investigate the implications of low-dimensional chaos on the ability of large recurrent neural networks to encode and discriminate time-dependent input signals, focusing on networks operating in a fluctuation-driven balanced state regime. We find that recurrent connections distribute signals throughout networks in ways that can enhance the classification power of the network, despite the chaotic behaviour that often results from recurrent connections.

back to overview