We systematically surveyed neurophysiological differentiation—a measure of the extent to which a population of neurons expresses a rich and varied repertoire of states, inspired by prior work on IIT—across layers and areas of mouse visual cortex in response to naturalistic movies and phase-scrambled controls.

The key finding was that naturalistic stimuli evoke more differentiated neural activity than scrambled stimuli, but only in specific populations: layer 2/3 excitatory neurons in the anterolateral and anteromedial visual areas. This effect was robustly modulated by arousal state: effect sizes showed strong correlations with locomotion and pupil diameter, suggesting that the difference in differentiation is more pronounced when animals are engaged.

Differentiation analysis represents an “inside-out” approach to neural activity, in the sense articulated by György Buzsáki (2019): rather than characterizing responses in terms of externally defined stimulus variables, it quantifies the intrinsic diversity of population dynamics. This contrasts with traditional “outside-in” methods such as decoding, which showed uniformly high performance across layers and areas and did not distinguish the specific populations highlighted by neurophysiological differentiation. In line with Romain Brette’s (2019) critique of the neural coding metaphor, these results underscore that decoding accuracy—defined relative to an experimenter’s variables and ideal observer assumptions—does not by itself establish functional or perceptual relevance. By characterizing population activity on its own dynamical terms, differentiation sidesteps these assumptions and may reveal which neural populations are signatures of functionally relevant dynamics—offering a new lens for understanding how neural activity relates to perception.