ICS Research seminar: "Multi-Transition Systems: An algorithmic perspective on neural spatial navigation"


"Multi-Transition Systems: An algorithmic perspective on neural spatial navigation" by Nicolai Waniek

Abstract: Spatial navigation is fundamental for animals and is attributed to place and grid cells in the rodent Hippocampal Formation. Found close to the apex of the cortical processing hierarchy, grid cells express prominent characteristics such as their eponymous hexagonal firing fields and a discrete modular organization along the dorso-ventral axis. Commonly believed to compute either path integration or localization, their true functional purpose remains unexplained. After presenting a brief overview of their characteristics and existing models, this talk introduces a novel algorithm and data structure centric perspective on place and grid cells. It is proposed that grid cells encode spatial transitions in spatio-temporal sequences and it is shown that their hexagonal responses arise when the number of available resources is minimized. Subsequently, their discrete modular organization is suggested to generate a scale-space for optimal transition look-ahead. Specifically, the scale-space forms a skip list, a well-known data structure that accelerates searches on linked lists. Important aspects of the presented model are its general capabilities, as well as fundamental links to other areas of research such as reinforcement learning, visual perception, and in particular multi-resolution path planning.

16. March 2018 at 10:30h, ICS Karlstraße 45, 2. floor in seminar room 2026