hide
Free keywords:
-
Abstract:
Ongoing neural activity unfolds across different timescales reflecting networks´ specialization for task-relevant computations. However, it is unknown whether these timescales can be flexibly modulated during trial-to-trial alternations of cognitive states (e.g., attention state) and what mechanisms can cause such modulations. We analyzed autocorrelations of population spiking activity recorded from individual cortical columns of the primate area V4 during a spatial attention task and a fixation task. We estimated timescales from auto-correlations using a novel method based on Approximate Bayesian Computations and applied a Bayesian model comparison to determine the number of timescales in neural activity. We found that at least two distinct timescales are present in both spontaneous and stimulus-driven activity. The slower timescale was significantly longer on trials when monkeys attended to the location of the receptive field of the recorded neurons than on control trials when monkeys attended to a different location. We hypothesized that the observed timescales emerge from the recurrent network dynamics shaped by the spatial connectivity structure.
We developed a network model consisting of binary units representing cortical minicolumns with local spatial connectivity among them. We found that the activity of model minicolumns exhibits two distinct timescales: A fast timescale induced by vertical recurrent excitation within a minicolumn and a slow timescale induced by horizontal interactions among minicolumns. The timescales depend on the network topology, and the slow timescale disappears in networks with random connectivity. We derived an analytical relationship between the timescales and connectivity parameters, enabling us to identify model parameters best matching the timescales in the data. The model indicates that modulation of timescales during attention arises from a slight increase in the efficacy of horizontal recurrent interactions. Our results suggest that multiple timescales in local neural dynamics emerge from the spatial network structure and can flexibly adapt to task demands.
