Abstract
When learning a complex task our nervous system self-organizes large groups of neurons into coherent dynamic activity patterns. During this, a network with multiple, simultaneously active, and computationally powerful cell assemblies is created. How such ordered structures are formed while preserving a rich diversity of neural dynamics needed for computation is still unknown. Here we show that the combination of synaptic plasticity with the slower process of synaptic scaling achieves (i) the formation of cell assemblies and (ii) enhances the diversity of neural dynamics facilitating the learning of complex calculations. Due to synaptic scaling the dynamics of different cell assemblies do not interfere with each other. As a consequence, this type of self-organization allows executing a difficult, six degrees of freedom, manipulation task with a robot where assemblies need to learn computing complex non-linear transforms and – for execution – must cooperate with each other without interference. This mechanism, thus, permits the self-organization of computationally powerful sub-structures in dynamic networks for behavior control.
Highlights
It is known, on the one hand, that networks can be trained to perform complex non-linear calculations[9,10,11], which could be used for motor control
The goal is to set up such a system that repeated stimulation should – by synaptic plasticity – trigger (i) the formation and outgrowth of cell assemblies while (ii) still keeping them functionally separate
The excitatory synaptic efficacy Wij between two units i and j has to be strong within an assembly and weak between them. These two requirements restrict the set of activity-dependent synaptic plasticity mechanisms (W ij = G[Fi, Fj, Wij ] with pre- and post-synaptic activities Fj and Fi and weights Wij) as they have to fulfill the following fixed-point condition of the synaptic dynamics: Wi⁎j ηFi κ
Summary
On the one hand, that networks can be trained to perform complex non-linear calculations[9,10,11], which could be used for motor control This requires that those networks produce a reservoir of rich, transient dynamics from which the desired outputs can be siphoned off. Self-organization of neurons into cell assemblies by the processes of synaptic plasticity induces ordered or even synchronized neuronal dynamics[1,6,22,23] This will reduce the dynamics of the network often to a degree that the required requisite variety for complex calculations cannot be provided by it any longer[15,24]. Trying to simultaneously create multiple assemblies will lead to the aforementioned catastrophic interference if one cannot prevent them from growing into each other
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have