Synaptic plasticity in biological neural networks

Aim

The brain’s unparalleled ability to learn and adapt is believed to rely largely on mechanisms of synaptic plasticity. Spike-timing-dependent plasticity (STDP) is one well-known mechanism that strengthens or weakens synapses depending on the firing rate and correlation structures of the connected neurons. We use a mathematical model of spiking neurons, the linear Poisson neuron, to investigate the impact of STDP in recurrently connected networks. We thus aim to predict what neural circuits in areas like the hippocampus or the cortex can learn, and how. In particular, we focus on how such circuits can become specialised in an unsupervised way when stimulated by sensory-like input pulse trains, in which information is encoded by firing rate and correlations.

Description

Mathematical techniques from stochastic analysis and dynamical systems allow us to determine the asymptotic activity state of our networks depending on the input characteristics and the learning parameters. These states have been related to properties of learning such as its stability and such as specialisation through symmetry breaking. We verify our analytical predictions using numerical simulation with simulation software we have developed (written in C++); a parallelised version to be used on a supercomputer is being developed.

Our results will lead to a better understanding of the information processing which takes place in the central nervous system. Such an understanding may allow us, for example, to fine tune the electric signals sent to the brain by electrodes, so as to make use of the natural plasticity of the brain.

People

Prof Tony Burkitt	Dr David Grayden
Dr Sean Byrnes	Ms Doreen Thomas
Dr Chris Trengove	Mr Matthieu Gilson