Weakly nonlinear interactions in an electrosensory cocktail party problem
Jan Benda - Eberhard Karls Universität TübingenJan Grewe - Eberhard Karls Universität Tübingen
Benjamin Lindner - Humboldt-Universität zu Berlin
The cocktail party problem emphasizes the amazing capabilities of the auditory system to separate one source of acoustic information from many other jamming signals. Particularly challenging for neural systems is the detection of faint signals close to threshold on a background of much stronger distractors. We propose to study the latter scenario both theoretically in populations of integrate-and-fire neurons and electrophysiologically as well as behaviorally in the active electrosensory system in two species of weakly electric fish. In field studies we recently observed male fish attacking a distant intruder while courting a close-by female. Here, the faint intruder signal is detected despite the presence of the more than a thousand fold stronger, ongoing female signal. The intruder and the female signal are both periodic but differ in their frequency. In contrast to the mammalian auditory system there are no frequency filter banks in the peripheral electrosensory system and all signals are processed by the same population of neurons. We hypothesize the enhancement of neuronal responses through weakly nonlinear interactions with the jamming signal, to be at the core of this impressive detection performance.We propose to characterize possible nonlinear interactions between two periodic signals in the first two stages of electrosensory processing in the gymnotiform fish Apteronotus leptorhynchus and A. albifrons. Guided by theoretical predictions we quantify the responses evoked by combinations of frequencies in electroreceptors and their targets, pyramidal cells that are organized in three maps. Across these maps the pyramidal cells differ in their cellular and functional properties and in their receptive field sizes. This rich parameter space allows to compare several evolutionary adaptations of a basic neural circuit design. In parallel we extend the theory of weakly nonlinear interactions to leaky integrate-and-fire neurons with adaptation currents. In addition we will consider the more natural case of a non-static signals. The electroreceptor population is heterogeneous with respect to baseline firing rates and response variability. We explore how heterogeneity might be opitmized for the representation of faint signals based on readout strategies of target neurons using various signal detection approaches. Additional operand conditioning experiments will quantify detection thresholds and integration times in controlled lab conditions and will provide important benchmarks for both the theoretical and electrophysiological work.Our work will provide solid and fundamental insights as to how weakly nonlinear interactions are involved in encoding faint signals in the presence of strong jamming stimuli. We will understand how evolutionary adaptations of a convergent network of neurons provide the basis for solving this difficult cocktail party problem.