In the rotation, you will develop a Python implementation of a simple neural model that learns to distinguish one class of inputs from another. It will thereby create a first encounter with biologically inspired models and machine learning.
In recent experiments it was found that the modification of synapses, which is broadly believed to underlie learning and memory in animal brains, is metabolically quite expensive. For instance, fruitflies that learn a simple association die about 20% earlier when feeding is stopped than fruitflies that did not learn the association. While the consequences of the metabolic cost of activity has been explored computationally a number of times and leads for instance to sparse coding and short-term synaptic depression, the metabolic cost of plasticity has not been considered before, yet from this and other data it appears to be an important design constraint.
MvR has started a programme to understand how the inclusion of metabolic costs affects the learning rules that neurons should normatively implement. In other words, how can one learn with minimal changes, or frugal plasticity.
In this particular project we will see how network of excitatory and inhibitory neurons adjust their connections so as to learn to store patterns while at the same time minimizing energy consumption. It has been suggested that in such network that there is a specific form of inhibitory plasticity that limits the neural activity. Interestingly, storing information in these networks will require a trade-off in the energy expended on modifying the synapses and the energy that is require to code information (which is for instance dependent on the relative contribution of excitation and inhibition in the network).
The expertise of Dr Ruediger Thul from the School of Mathematics on calcium dynamics is crucial as the exact calcium dynamics determines the plasticity of the synapse and thereby the metabolic cost. For instance we hypothesize that under conditions of low energy the calcium influx is restricted so as to save energy. Secondly, the calcium dynamics might affect the synaptic consolidation of synapses, which has been suggested to be a particularly metabolically expensive process.
As the project complements a recently awarded Leverhulme grant, you will be able to gain extra support and will have the opportunity to collaborate with the postdoc employed on that grant.
The University of NottinghamUniversity Park
Nottingham, NG7 2RD
Tel: +44 (0) 115 8466946
Connect with the University of Nottingham through social media and our blogs.
Campus maps | More contact information | Jobs
Browser does not support script.