Oscillations and Gain Control in Sensory Systems

• Main Author: Payeur, Alexandre
• Other Authors: Longtin, André
• Language: en
• Published: Université d'Ottawa / University of Ottawa 2016
• Subjects: Sensory systems; Neural networks; Oscillations; Gain control; Linear response theory
• Online Access: http://hdl.handle.net/10393/34205; http://dx.doi.org/10.20381/ruor-5466

Sensory neurons assemble to form networks that process inputs coming from the senses. Through synaptic connections neurons interact and create complex dynamical states in response to these inputs. Networks with different connectivity patterns are thought to display different states and therefore subserve different computational goals. In this thesis, we mainly study brain rhythms, a dynamical state that occurs in various neural structures. Rhythms are emergent oscillations that typically occur in homogeneous recurrent networks, whose neurons have identical properties and are densely interconnected. Many sensory systems comprise neurons with opposite ON and OFF responses to inputs. We show that homogenous recurrent networks fail to sustain rhythms when ON and OFF neurons are present in equal proportions. This happens even when the network is subjected to spatially correlated inputs, which are known to promote synchronized oscillations. In this context, we adapted the so-called linear response theory to include networks containing ON and OFF neurons with different intrinsic properties. In this asymmetric case, oscillations can be recovered. A simpler approach is to segregate the ON and OFF populations, thus producing two oscillating subnetworks. The dynamics of purely feedforward networks are studied next. These networks are composed of two or more populations. The populations are connected in a serial fashion, but neurons are unconnected within the populations. This connectivity scheme is drastically different from the fully recurrent network. Yet, this network is shown to display oscillatorylike properties when subjected to spatially correlated stimulation under certain conditions. We also find that this network can implement various types of gain control, depending on the noise in the system and the strength of synaptic interactions. These results establish some unexpected links between feedforward and recurrent networks. Along the way, we apply our results and conclusions to a well-characterized sensory network, the electrosensory system of weakly electric fish.