Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster F in Poster Session F: Thursday, August 6, 1:45 – 3:30 pm, Kimmel Center, Shorin & Rosenthal Rooms

LieCAN: Neural attractor networks that perform real-world pose representation and path integration

Deven Shidfar¹, Khanh Dao Duc¹, Manu Madhav¹; ¹University of British Columbia

Presenter: Deven Shidfar

The hippocampal cognitive map represents the pose of an animal (𝐱,𝐲,θ) in allocentric coordinates. However, self-motion-based velocity cues are sensed in egocentric coordinates. While continuous attractor networks (CANs) provide a framework for representing and updating pose, these models typically transform egocentric velocity signals into allocentric at every time step, introducing a computation that may be costly and noise sensitive. We propose LieCAN, an architecture that utilizes the group structure of pose representations, in which the egocentric-allocentric transformation is embedded in the recurrent connectivity. The network directly integrates egocentric velocity inputs without requiring an explicit coordinate transformation at each step. Our results suggest that LieCAN path integrates more accurately compared to a CAN that requires explicit transformation. Biologically plausible navigation circuits may benefit from hard-coding the coupling between egocentric and allocentric coordinates directly into their connectivity, motivating future experiments to test whether similar computations are implemented in the brain.

Topic Area: Methods, Tools, Theory & Neural Coding