Targeted comodulation supports flexible and accurate decoding in V1

Abstract

Sensory-guided behavior requires reliable encoding of stimulus information in neural populations, and task-specific readout through a selective combination of these responses. The former has been the topic of extensive study, but the means by which the brain achieves the latter remain poorly understood, especially when adapting to changing task demands. Here we introduce a novel theory for adaptive sensory processing based on functionally-targeted stochastic modulation. We find that responses of neurons in area V1 of monkeys performing a visual orientation discrimination task exhibit low-dimensional, rapidly fluctuating gain modulation, which is stronger in neurons that are most informative for the current behavioral task. We propose that this modulation serves as a label that supports adaptive downstream readout. Our theoretical and experimental results show that V1 modulation can be used to decode from neural activity after only a small number of training trials, consistent with observed behavior. In a hierarchical visual neural network model, the modulator-induced labels are learned quickly and accompany task information across several processing stages to finally guide decisions. Consistent with this model, we find that the V1 modulatory signal is also present in the activity of simultaneously recorded MT units, and that its label of task information is preserved. Our findings provide evidence for a novel mechanism for task-adaptive information routing in the brain, through targeted co-modulation.