Abstract
Stimulus-independent spike-rate variability tends to be weakly correlated between neuronal pairs throughout sensory cortex. It is frequently suggested that these "noise" correlations (rsc) reflect the stochastic nature of sensory input pathways, from which it follows they place hard limits on the fidelity of sensory encoding. For an optimal linear decoder, correlations limit information when they follow a particular relationship with neuronal preferences (so-called "differential" correlation structure). However, rsc may also reflect shared input from signals that are centrally generated and potentially under voluntary control. If so, its impact on sensory encoding is unclear. We sought to directly test whether the structure of noise correlations in visual cortex reflects signals of central origin during perceptual decision making, by measuring rsc in contexts where task instruction changed but retinal input was fixed. We recorded population spiking activity in primary visual cortex (V1) of rhesus monkeys, while the subjects performed a coarse orientation discrimination task using filtered noise. The discriminanda were fixed in a given recording session but were varied between sessions. We found that rsc structure changed with the task even on zero-signal trials which were identical for all sets of discriminanda, indicating a central origin. Specifically, pairs of neurons which preferred the same orientation discriminandum were more highly correlated than pairs preferring opposite discriminanda, while pairs not well tuned for the task showed no modulation. At first sight, these changes appear to have a detrimental impact on the information capacity of the population, since they are nearly "differential" in structure. However, if downstream areas can distinguish inputs of central and peripheral origin, then the observed rsc structure no longer constrains the information limit. More complex models of the relationship between feedforward and feedback pathways in sensory processing are needed to determine when noise correlations between sensory neurons are truly information-limiting
Meeting abstract presented at VSS 2016