Conference on Cognitive Computational Neuroscience

Spontaneous activity is ubiquitous in the brain, but relatively little is known about its temporal dynamics. We ask how spontaneous activity traverses its high-dimensional space on different time scales, when unfolding in time. We seek to distinguish between random sampling; slower than random sampling and faster than random sampling. We study these scenarios with the correlation between subsequent patterns of spontaneous activity as well as the growth of linear dimensionality over time. We find that the temporal correlation and evolution of dimensionality are indistinguishable from randomly shuffled time series in i) spike patterns recorded in populations of single neurons in macaque V1, ii) widefield calcium imaging data recorded in ferret V1, and iii) human whole brain fMRI. However, we do find that consecutive vectors, tangent to the activity trajectory in lower-dimensional spaces resemble each other. These consistent temporal directions hint at persistent activity propagation. Thus, we conclude that on time scales of a few seconds, spontaneous activity samples its activity space as rapidly as a process of random sampling, even in the presence of persistent directions.