Neural geometry in the human hippocampus enables generalization across spatial position and gaze

A neuroscience collaboration led by researchers from Columbia University and University of Pennsylvania has published groundbreaking research on how the human hippocampus organizes spatial information. The study, involving 15 authors including Benjamin Hayden and Joshua Jacobs, recorded from hippocampal neurons while participants performed a joystick-controlled virtual prey pursuit task with multiple moving agents. They discovered that neurons exhibit mixed selective responses that simultaneously map positions of self, prey, predator, and gaze direction, with these different codes occupying mostly orthogonal subspaces within the neural population.

The key finding reveals that these subspaces maintain a geometric structure that allows them to be aligned through simple linear transformations. This neural architecture enables what researchers call 'generalization across spatial maps'—a linear rule learned for tracking one agent can transfer to another. The hippocampus appears to structure spatial knowledge as a family of geometrically related manifolds that can be flexibly aligned to different agents and viewpoints. This explains how humans can track multiple entities in complex environments without cognitive confusion while maintaining the ability to abstract spatial relationships across different perspectives.