Stochastic resonance in the impact of afferent sensory noise on grid-patterned firing and path integration in a continuous attractor network.

The continuous attractor network (CAN) model explains grid-patterned firing and path integration in the entorhinal cortex, yet the impact of sensory noise in velocity inputs remains unexplored. In addressing this, we introduced varying levels of noise to the velocity inputs impinging on a 2D CAN model driven a virtual animal traversing an arena. We estimated animal position from network activity and quantified position accuracy as the difference between real and estimated positions. We performed all simulations using several trajectories, as grid scores and position accuracy showed pronounced trajectory-to-trajectory variability even without noise. We found that low levels of sensory noise were beneficial to grid-field formation, particularly for trajectories that failed to generate grid patterns in noise-free conditions. For trajectories exhibiting grid-patterned activity without noise, low levels of noise improved position estimation accuracy. In contrast, high levels of sensory noise impaired position estimates and grid-patterned activity. These results demonstrate stochastic resonance in a 2D CAN model, where an optimal level of sensory noise enabled grid-patterned activity and enhanced position accuracy. Motivated by the proposed error-correcting role of border cells, we introduced north and east border cells that were connected to grid cells based on co-activity patterns. Interestingly, while border inputs enabled grid field formation in cases where grid patterns were previously absent, their effect on position accuracy was marginal. Together, our analyses suggest that biological CANs could evolve to yield optimal performance in the presence of noise, which could serve as a stabilizing factor yielding functional robustness through stochastic resonance.