Cortical coding favors high-dimensional diversity over fixed categories

Across ~14,000 neurons in 43 mouse cortical regions, the study shows cortex-wide representations are organized categorically, but within individual areas responses are highly diverse and non-categorical. This diversity creates high-dimensional neural geometry that enables linear readouts to separate many conditions, with maximal separability emerging when information is accounted for in each area. The results imply a computational regime that prioritizes diversity over fixed categories, and reveal a gradient of increasing representational dimensionality along the cortical hierarchy closely tied to anatomy and connectivity.