Panteleimon Vafeidis: Learning of a path-integrating circuit

BCCN Berlin and TU Berlin

Abstract

Head direction cells are neurons that respond selectively to a specific head orientation in space, and along with place and grid cells, they are an important component of the neural circuitry coding for space in the hippocampal formation. Head direction cells are organized in a recurrent network, where the current head orientation is represented by a localized bump of neuronal activity. Path integration in the context of head direction cells is moving this bump of activity around the network respecting the animal's current head orientation, while using only self-motion cues. The one-dimensional ring attractor has been a popular approach to model the network of head direction cells, and one of the most long-lived attractor network models in Computational Neuroscience. Furthermore, this model has recently received experimental support in the fly's head direction system. However, it remains to date unclear how a path integrating circuit of head direction cells could self-organize through synaptic plasticity. Previous models showed circuits that either needed parameter tuning after training, or used biologically implausible learning rules. Here we propose a model that self-organizes to achieve the required goal, i.e. error-free path integration in the absence of visual cues, while remaining biologically plausible. Finally, we lay the path to adopt the same approach to a well-characterized head direction circuit in the fly's brain.

Master thesis defense

Organized by

Susanne Schreiber, Richard Kempter

Location: ITB seminar room 012, Building 4, Philippstr. 13

Go back