Neuronal assembly formation and non-random recurrent connectivity induced by homeostatic structural plasticity
Time: Mon 2020-06-08 13.00
Doctoral student: Julia Gallinaro , Datavetenskap, Bernstein Center Freiburg and Faculty of Biology, University of Freiburg, Germany
Opponent: Wulfram Gerstner,
Supervisor: Pawel Herman, Beräkningsvetenskap och beräkningsteknik (CST); Alexander Kozlov, Skolan för elektroteknik och datavetenskap (EECS); Stefan Rotter, Bernstein Center Freiburg and Faculty of Biology, University of Freiburg, Germany
Abstract
Plasticity is usually classified into two distinct categories: Hebbian or homeostatic. Hebbian is driven by correlation in the activity of neurons, while homeostatic relies on a negative feedback signal to control neuronal activity. Since correlated activity leads to strengthened synaptic contacts and formation of cell assemblies, Hebbian plasticity is considered to be the basis of learning and memory. Stronger synapses, on the other hand, promote stronger correlation. This positive feedback loop can lead to instability and homeostatic plasticity is thought to play a role of stabilization. The experimentally observed time scales of homeostatic plasticity, however, are too slow to compensate for the fast Hebbian changes. Therefore, the exact way multiple types of plasticity interact in the brain remains to be elucidated. In this thesis, we will show that homeostatic plasticity can also have interesting effects on network structure. We will show that homeostatic structural plasticity has a Hebbian effect on the network level, and it comprises two separate time scales, a faster for learning and a slower for forgetting. Using a model of classical conditioning task, we will show that this rule can perform pattern completion, and that network response upon stimulation is gradual, reflecting the strength of the memory. Furthermore, we will show that networks grown with homeostatic structural plasticity and a broad distribution of target rates exhibit non-random features similar to some of those found in cortical networks. These include a broad distribution of in- and outdegrees, an over-abundance of bidirectional motifs and scaling of synaptic weights with the number of presynaptic partners. Overall, we will use simulations of spiking neural networks and mathematical tools to show that there is more to homeostatic plasticity than just controlling network stability. It remains an open question, however, the extent to which homeostatic plasticity can be accounted for structural features found in the brain.