Complex/Dynamical Systems Seminar - Joel Zylberberg

Maintenance and dynamics of working memory networks

Working memory is a key cognitive function that requires information about a task, or about the external world, to persist within the brain for many tens of seconds. This talk will address two puzzles about the mechanisms of working memory. First, experiments show that neurons do not behave deterministically, but rather show high levels of noise. During a memory task, noise from different sources has ample time to accumulate, potentially degrading the memory representation. In my talk, I will explain how the brain solves this problem, and generates robust memory function in systems of noisy components. Next, we will turn our attention to the problem of timescales: while representations persist for tens of seconds during working memory tasks, the activities of individual neurons change over timescales of tens of milliseconds. For networks of neurons, the timescales can be much longer than those of single neurons, but the network connectivity must be carefully arranged for this to work. This “fine tuning” is problematic because connectivity in neural networks changes over time, and so even if the network starts out finely tuned, it will very quickly lose that tuning, unless the changes in connectivity support the tuning. In my talk, I will show that synaptic plasticity — which changes the connectivity of brain networks - can dynamically re-tune memory networks, and make some predictions for the types of mechanistic perturbations that should degrade working memory function.