Decrease in within-trial variability contributes to a decrease in across-trial variability of neural firing in the primate cortex during neural computations

Abstract

The conventional approach to understanding neural responses underlying complex computations is to study across-trial averages of repeatedly performed computations from single neurons. When a brain region performs complex computations, such as processing stimulus related information or motor planning, it has been repeatedly shown through measures such as the Fano factor (FF) that neural variability across trials in the network decreases. However, in reality, multiple neurons contribute to a common computation on a single trial, rather than a single neuron contributing to a computation on multiple trials. Therefore, on individual trials the concept of FF loses significance. In this study, we extended previous work using measures of variability that are confined to a single trial and found that neurons perform a common computation, they briefly fire with increased regularity in spike timings, with similar inter-spike interval durations. We propose that this decrease in within-trial variability in neural spiking contributes to a decrease in across-trial variability in neural firing rates during network level computations. We confirmed our hypothesis by testing it on the activity of frontal eye field neurons recorded as two monkeys performed a memory-guided saccade task, and also on simulated spike trains. Furthermore, this phenomenon also has important behavioral correlates: the reaction time of the animal was faster when the within-trial variability was lower. We show that a decrease in within-trial variability is linked to a decrease in across-trial variability in neural response and indicates stationarity of neural network variability across time.\n\nNew & NoteworthyDuring computations, neural variability across trials decreases. In reality, multiple neurons contribute to a common computation on a single trial, rather than a single neuron contributing to a computation on multiple trials. We found that when a network of neurons performs a common computation, they briefly fire with increased regularity in spike timings. We propose that this decrease in within-trial variability in neural spiking contributes to a decrease in the observed across-trial variability.