Poster at Cosyne on neural timescales across cortical hierarchy

I’m presenting a poster at the 2013 Cosyne meeting, on Saturday March 2. My poster is titled: Hierarchy of intrinsic timescales across primate cortex. In collaboration with electrophysiologists, this project combines data analysis and computational modeling to explore the neural circuit basis of specialized function across cortical areas. My abstract is below:

Hierarchy of intrinsic timescales across primate cortex

John D. Murray, Alberto Bernacchia, Tatiana Pasternak, Camillo Padoa-Schioppa, Daeyeol Lee, Xiao-Jing Wang

Primate cortex is hierarchically organized, and different cortical areas appear specialized for diverse computations. However, the neural circuit basis underlying these areal specializations remains an open question. One hypothesis is that differences in local circuit properties across areas may be responsible for this specialization. We hypothesize that, at the physiological level, these differences can be detected in terms of differential timescales of neural dynamics.

To test this hypothesis, we studied temporal autocorrelation of single-neuron spike trains in multiple areas, across sensory, parietal, and prefrontal cortex, recorded in monkeys performing cognitive tasks. We focused on activity during the task’s fixation period to isolate internal dynamics and facilitate comparison across different datasets.

In a given cortical area, decay of autocorrelation is well described by a characteristic timescale, which reflects intrinsic firing rate fluctuations within single trials. Across areas, timescales follow a hierarchical ordering, with sensory and prefrontal areas exhibiting short and long timescales, respectively, spanning an order of magnitude. The hierarchy of intrinsic timescales correlates with hierarchies derived from long-range anatomical projections, linking physiological and anatomical measures.

The autocorrelation decay also exhibits an offset, which may reflect rate fluctuations on slower timescales. In particular, the offset could reflect the strength of across-trial memory encoded in firing activity. To test this possibility, we used a decision-making task that demands across-trial memory, in which neurons exhibit a characteristic timescale for memory of past rewards. In line with this interpretation, we find that autocorrelation offset correlates with the timescale of reward memory.

To explore potential mechanisms, we studied a spiking neural circuit model. We find that strengthening recurrent structure in the network increases the intrinsic timescale, within the range observed experimentally. We propose the hierarchical gradient of intrinsic timescales across cortical areas reflects specialization of local circuit properties for diverse computations.