Emergent Navigational Computations in a Self-Organizing Attractor Model

While continuous attractor neural networks (CANNs) model spatial representations and prospective coding in navigational circuits, joining biologically plausible learning with diverse navigational phenomena remains challenging. We developed a self-organizing firing rate model where a neuronal layer, exhibiting competitive dynamics via adaptation and global inhibition, spontaneously learns to represent space through local Hebbian plasticity. Crucially, this framework captures three complex findings. First, population activity naturally anticipates stimulus positions, matching entorhinal predictive coding. Second, preliminary results suggest recurrent plasticity offers a mechanistic explanation for dynamic speed recalibration in path integration, reflecting rapid hippocampal gain adjustments to cue conflicts. Finally, it addresses novel 3D navigation data : on inclined surfaces, entorhinal grid maps undergo a small, stationary upward shift along the slope axis, preserving overall symmetry. By increasing the synaptic contribution of downhill-oriented conjunctive cells, our network accurately reproduces this shift in omnidirectional cells. Thus, a single framework of adaptation, competition, and synaptic plasticity is sufficient to self-organize circuitry for predictive coding and terrain-dependent mapping, while providing analytical tractability to explore dynamic speed recalibration.