Grid Cell as Interface Representation between Animal Brains and Artificial Neural Networks

With recent breakthroughs in the electrode-based brain-computer-interface technology, cognition enhancement becomes increasingly feasible by applying external stimuli to neurons in the brain. An essential stepstone for this objective is to develop compatible representations between artificial and biological neural networks. We propose that grid cells can serve as such an interfacing representation. First, we empirically proved that DNNs replicated the grid cell behaviors on non-spatial cognitive tasks. Second, we developed theoretical analysis on the formation of hexagonal activation pattern of grid cells, which is independent of physical implementations of neural networks. Third, we performed extensive experiments to validate our proposal.