Information transmission vs processing in computer models of neocortical columns

Neuronal networks can produce many different patterns of activity that can be characterized in terms of the level of inter-neuronal correlation. This level determines the degrees of freedom the network has to process, encode, and transmit information. At one extreme, neurons fire asynchronously, with each being nearly independent in a loosely coupled network. In this case the neurons are free to fire independently and the network can nominally transmit more information, since there is a larger ensemble of possible network firing states. However, the network is then not processing information; information is not being transformed but only transmitted. Presumably there are ways the network design can optimize this transmission/processing trade-off depending on the needs of a particular brain area. We designed computer simulations of neuronal networks of layered cortical columns to produce either (a) transmission or (b) processing networks. Randomly-timed sub-threshold synaptic inputs were provided throughout the simulation to provide baseline activity. The network responded in dramatically different ways, depending on the pattern, frequency, and weights of random stimuli provided as well as on the internal network properties. We used normalized transfer entropy and Kendall's tau correlation to measure the degree to which the network followed the inputs. In addition we measured the correlation of input to output frequencies. With loose internal coupling, the network followed the input with a high degree of correlation, and high transfer entropy from the inputs to the neuronal spike outputs. Strengthening the network's internal connectivity permitted gradual reduction in the degree of passive response with increase in intrinsic activity, including higher correlation and transfer entropy between the cells within the network. The inputs were also selectively applied to different layers to see how each transmitted/processed the inputs. We then measured the correlation and information transfer between and within layers in the cortical column as well as between the inputs and outputs at different layers. The gains of synaptic weights were also modified to test for optimal signal detection/propagation from input to output layers. Low internal inhibition predisposed to avalanches of synchronized activity in cell populations, while high internal inhibition led to synchronized oscillations in inhibitory interneuron populations with minimal activation from excitatory cells.