The dynamic connectome

Abstract

When trying to gain intuitions about the computations implemented in neural circuits, we often use comparisons with electronic circuits. However, one fundamental difference to hard-wired electronic circuits is that the structure of neural circuits undergoes constant remodeling. Here, we discuss recent findings highlighting the dynamic nature of neural circuits and the underlying mechanisms. The dynamics of neural circuits follows rules that explain steady state statistics of synaptic properties observed at a single time point. Interestingly, these rules allow the prediction of future network states and extend the insights gained from serial sectioning electron microscopy of brain samples, which inherently provides information from only a single time point. We argue how the connectome’s dynamic nature can be reconciled with stable functioning and long-termmemory storage and how it may even benefit learning.

Dendritic Computing: Branching Deeper into Machine Learning

“…While most of the brain volume is consumed by axons, the neurons in the brain are still extremely sparsely connected. Modern imaging techniques have shown that this connectivity matrix is highly dynamic, with synaptic connections being established and removed on the time scale of hours and days (Holtmaat et al, 2005;Stettler et al, 2006;Kasai et al, 2010;Loewenstein et al, 2011;Rumpel and Triesch, 2016). It was hypothesized that these dynamics are used by the brain to optimize network function under severe constraints on connectivity (Kappel et al, 2015b(Kappel et al, , 2018.…”

Drifting Assemblies for Persistent Memory

“…Organisms change over time, on many dierent levels. This holds in particular for the synapses in neural networks [1]: They change their impact and also appear and vanish. On the one hand, this weight and structural plasticity is activity dependent.…”