Item

Abstract

Ongoing cortical dynamics unfold across different temporal scales. These timescales reflect the network’s specialization
for task-relevant computations. However, it is unknown how different timescales emerge from the spatial
network structure and whether they can be flexibly modulated by cognitive demands, e.g., during attention. We
developed a network model which consists of binary units representing local neural populations (mini-columns)
with spatially structured connections among them. We find that activity of the mini-columns exhibits two distinct
timescales arising from the network dynamics. The first timescale is induced by recurrent excitation within
a mini-column (vertical connectivity), and the second timescale is induced by interactions among mini-columns
(horizontal connectivity). The timescales depend on the network topology, and the second timescale disappears
in networks with random connectivity.
To test model predictions, we analyzed spiking activity recorded from single cortical columns in the primate areas
V1 and V4 during an attention task. We developed a novel method based on adaptive Approximate Bayesian
Computations, which estimates the timescales from spiking activity and overcomes statistical biases due to finite
sample size. We observed two timescales in both V1 and V4 population dynamics. Both timescales were
significantly longer in V4 than in V1, which is explained by our model based on differences between V1 and V4
network properties. Moreover, the V1 and V4 timescales were longer when attention was directed toward neurons’
receptive-fields. This result reveals how ongoing network dynamics is influenced during top-down attention even
without measurable modulations of firing rates in the absence of visual stimuli. Based on our model, modulation of
timescales arises from an increase in efficacy of vertical connections and a slight suppression of horizontal interactions.
Our results suggest that timescales of local neural dynamics emerge from the spatial network structure
and can flexibly change due to top-down influences according to task demands.