Uncertain Heterogenous Algorithmic Teamwork

Lead Research Organisation: University of Liverpool
Department Name: Electrical Engineering and Electronics

Abstract

This project is focused on showing that raw aggregated capability can outperform carefully constructed co-ordination. More specifically, while dedicated high-performance computing resources provide a carefully constructed homogeneous environment that can make best use of the available hardware, there are settings where the availability of vast quantities of computational hardware should more than make up for the disparate connectivity and capability of the hardware. Recent collaboration between the University of Liverpool and IBM Research has developed numerical Bayesian techniques that exploit homogeneous super-computing hardware to outperform algorithms that are, by default, configured to make use of a single processing core. These techniques pave the way for a generic solution to the problem of performing algorithmic teamwork in the context of data science.

This PhD will investigate whether it is possible to adapt the pre-defined divide-and-conquer algorithm at the heart of the aforementioned numerical Bayesian techniques to adapt to and operate effectively within an uncertain computational environment. This will involve the development of fast distributed algorithms to identify and instantiate a divide-and-conquer architecture that is near-optimal given the available resources. The aim of the project is to develop the infrastructure that makes it possible for vast heterogeneous compute resources to operate effectively in a team. If successful, the aim is to use spare computer infrastructure available globally to deploy algorithms based on teamwork to answer a fundamental societal question which will soon be identified.

Publications

10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/S023445/1 01/04/2019 30/09/2027
2270559 Studentship EP/S023445/1 01/10/2019 30/09/2023 Matthew Carter