Funded Computer Science PhD Studentship Big Data Analytics: Visualising high-dimensional cost function landscapes

University of Exeter – College of Engineering, Mathematics and Physical Sciences

Primary Supervisor: Dr Jonathan Fieldsend

Secondary Supervisor: Dr Ozgur Akman

Project Description:
With the vast growth in scientific data, data visualisation methods have become ever more important. These are crucial to both bridge the gap between specialists and non-specialists (to aid the explanation of science and results), and also to investigate and probe the relationships within data (leading to new knowledge and discoveries).

One area where such visualisation is important, is when undertaking and examining the results of an optimisation. For instance, visualisation of the fitness landscapes relating designs to their corresponding quality, and broader differences between design regions is useful, but difficult, as the data may naturally live in a high number of dimensions.

This College-funded PhD project, is closely aligned to the EPSRC-funded project, EP/N017846/1: The Parameter Optimisation Problem: Addressing a Key Challenge in Computational Systems Biology. It is concerned with the investigation and development of novel visualisation methods for the data generated from work on optimising gene network models, with circadian clocks as a prototypical example, although some aspects will be applicable to a broader class of dynamic network models.

A key visualisation goal will be conveying mode distributions and magnitudes via lower-dimensional representations of much higher-dimensional search spaces. A related objective will be the development of metrics for quantifying how the topological structure of the parameter space is modified by the projection onto the visualisation space. i.e. we will need to incorporate information on the distances in the original space between modal regions – as well as the mode magnitude and volumes – into the visualisation method. This will provide insight into the information provided by different experimental datasets, and yield a quantitative basis through which to potentially reduce the number of objectives down to a maximally informative subset, thereby reducing the computational complexity of the optimisation.

Funding Minimum
3.5 year Studentship: Tuition fees UK/EU and an annual maintenance allowance at current research council rate (£14,553 for 2017-18)


Leave a Reply