Director of studies: Dr Alma Rahat
2nd supervisor: Professor Roman Borisyuk
Applications are invited for a three-year PhD studentship. The studentship will start on 1 October 2019.
The PhD student will join an interdisciplinary team of experts in optimisation and computational neuroscience to answer an extremely important question: how does the neural connectivity determine behaviours? To answer this question, the student will design computational models of neural networks, and use advanced machine learning and optimisation techniques to find the correspondence between connectivity and functionality.
Information processing in the brain, and ultimately observed behaviour, are based on the communication between spiking neurons that are embedded in a network of synaptic connections. In nature, different individual organisms usually have different connectivity (represented by a graph of interconnected nodes) that can produce the same behaviour. It is therefore not obvious how to locate optimal neural connectivity to match a desirable behaviour. In this exciting project, we aim to develop a novel approach towards solving this optimisation problem.
The connectivity graph can be described in terms of connection probabilities, which should be optimised to generate a target behaviour. This is a very important problem, but the computational complexity renders traditional optimisation algorithms impractical. Thus, we expect Bayesian optimisation to be an efficient approach in this context. However, this method must be adapted for graph-based problems. In this project, we will explore various uncertainty quantification methods for graph connectivity to enable the use of Bayesian optimisation. To demonstrate the performance and test the efficacy of our methods, we will study the motor behaviours of a simple animal for which neurobiological data and effective models are available.