Millimeter-scale selective amplification in the developing visual cortex

Cortical processing is believed to involve the selective amplification of specific inputs by recurrent networks. But whether selective amplification operates within the millimeter-scale networks underlying perception and behavior remains unclear. Here, we combine patterned optogenetic stimulation--informed by computational models--with calcium imaging in immature ferret visual cortex to show that cortical networks are preferentially activated by inputs aligned to endogenous recurrent subnetworks. Dominant subnetworks can be inferred from spontaneous activity, where response reliability and specificity depend on the overlap of input with these dominant modes. Well-aligned inputs are selectively amplified, evoking more reliable and specific responses that exhibit greater spatiotemporal stability than misaligned inputs. This preferential amplification suggests early cortical networks leverage endogenous dynamics to stabilize and refine precise sensory representations over development.