David Aristoff's homepage Title: Stochastic neural dynamics of working memory

Abstract: To navigate the world, organisms often store information temporarily and retrieve it a few moments later. Such short term storage can be accomplished solely by neural activity, rather than the network plasticity required for long term memory. Neuronal activity underlying working memory can be modeled by networks with local excitation and broad inhibition, which support persistent bumps of activity. We consider a specific task where a subject must remember a single analog variable. In this case, fluctuations cause the bump to wander, degrading the fidelity of the memory. Considering the effects of spatially heterogeneous coupling, this diffusion can be reduced, and the resulting dynamics can be approximated by a simplified potential well model. This mathematical framework can be extended to the case of multi-layer models and delayed coupling. Ultimately, we find that short term memory is improved in networks that consider these more detailed features of neural architecture.