Talk by Dr. David Kappel (CSN Virtual Seminar)

We hereby announce the next talk in the 'CSN Virtual Seminar':

Understanding synaptic dynamics: Structural plasticity and stochastic release

by Dr. David Kappel, Ruhr-Universität Bochum

Abstract

Biological synapses show a high level of seemingly random and very dynamic rewiring and substantial trial-by-trial variability for stochastic release events, suggesting that synapses are highly unreliable and a major source of noise in the brain. These findings are surprising, given how costly synapses are to maintain for an organism. In this talk I will discuss two approaches to uncover the principles by which biological neural circuits reliably perform complex learning tasks despite - or even with the help of - synaptic variability. The first approach incorporates experimental results on the complexity and variability of biological synapses and their rewiring dynamics, and utilizes these mechanisms for automatic self-organization of neural circuits. This model allows us to better understand the high variability of synaptic strengths in biology and also provides the basis for an implementation of efficient learning in neuromorphic hardware and bio-inspired learning algorithms. In ongoing work we augment the synaptic dynamics to selectively stabilize synapses and to protect memory items from interference. The comprehensive model links the complex biological mechanisms that govern synaptic efficacy changes to state-of-the-art machine learning. The second approach targets the surprisingly high experimentally found trial-by-trial variability of synaptic release. Here we pursue a top-down approach that describes synaptic dynamics through the goal to minimize the prediction error of future post-synaptic spike sequences, and use this modeling framework to derive learning rules for a recurrent spiking network model. The emerging learning rules are local, resemble experimentally found plasticity mechanisms and promote the formation of stable neural assembly sequences that become active in synchrony with afferent inputs. These results provide new insights to uncover the role of synaptic noise in enabling recurrent spiking networks to self-organize.