The Effects of Scaling of Columnar Cell Population Sizes in a Spiking Attractor Network Model of Working Memory
CBN (Computational Biology and Neurocomputing) seminars
Friday 18 September 2015
to 11:00 at
Experimental neuroscientific research has in the past decades identified the elevated activity of distinct populations of recurrently connected neurons in the brain as the biological correlate of working memory activation. The theoretical foundation of this progress has predominantly relied on the idea that information is encoded in the pooled response of recurrently con- nected cell assemblies operating with attractor dynamics. The discovery of columnar organization in neocortex has further strengthened this theory. Yet, the composition, and even existence, of such minimal units of computation in the brain remains elusive. In this study, we assess the concept of columnar cell assemblies as computational units in the framework of a spiking attractor network model of prefrontal, discrete working memory, by studying the robustness of spike dynamics and pattern completion functionality to scaling of minicolumnar cell populations. The results suggest that each such minicolumn requires at least 25–30 neurons in order to maintain irregular and sparse firing, as well as low noise correlations. In addition, preliminary trials demonstrate that too sparsely populated units destabilize temporal dynamics in the idling and coding state, thus prohibiting bistability. These findings affirm the necessity of ensemble coding in the network model.