Do you enjoy working with probabilities, data, and algorithms? Are you interested in the theory of causality? Do you want to improve the methods we use to discover cause-and-effect relationships from data and knowledge? Are you interested in algorithms that discover Bayesian Network (BN) graphs for causal inference, and Bayesian Decision Network (BDN) graphs for optimal decision making; i.e., maximising utility and minimising risk as in game theory?
The PhD student will specialise in the theory and application of Bayesian Networks, with a focus on structure learning; i.e., the automated discovery of the graphical structure/network. The PhD studentship is part of the EPSRC project on Bayesian Artificial Intelligence for Decision Making under Uncertainty:
Scientific research is heavily driven by interest in discovering, assessing, and modelling cause-and-effect relationships as guides for action. Much of the research in discovering relationships between information is based on methods which focus on maximising the predictive accuracy of a target factor of interest from a set of other related factors. However, the best predictors of the target factor are often not its causes and hence, the motto "association does not imply causation". Although the distinction between association and causation is nowadays better understood, what has changed over the past few decades is mostly the way by which the results are stated rather than the way they are generated. Bayesian Networks (BNs) offer a framework for modelling relationships between information under causal or influential assumptions, which makes them suitable for modelling real-world situations where we seek to simulate the impact of various interventions. BNs are also widely recognised as the most appropriate method to model uncertainty in situations where data are limited but where human domain experts have a good understanding of the underlying causal mechanisms and/or real-world facts. Despite these benefits, a BN model alone is incapable of determining the optimal decision path for a given problem. To achieve this, a BN needs to be extended to a Bayesian Decision Network (BDN), also known as an Influence Diagram (ID). In brief, BDNs are BNs augmented with additional functionality and knowledge-based assumptions to support the representation of decisions and associated utilities that a decision maker would like to minimise or maximise . As a result, BDNs are suitable for modelling real-world situations where we seek to discover the optimal decision path to maximise utilities of interest and minimise undesirable risk. Because BNs come from statistical and computing sciences, and whereas BDNs come mainly from decision theory introduced in economics, research works between these two fields only occasionally extend from one field to another. As a result, it is fair to say that the landscape of these approaches has matured rather incoherently between these two fields of research. It is possible to develop a new generation of algorithms and methods to improve the way we 'construct' BDNs. The overall goal of the project is to develop an open-source software that will enable end-users, who may be domain experts and not statisticians, mathematicians, or computer scientists, to quickly and efficiently generate BDNs for optimal real-world decision-making. The proposed system will allow users to incorporate their prior knowledge for information fusion with data, along with relevant decision support requirements for intervention and risk management, but will avoid the levels of manual construction currently required when building BDNs. The system will be evaluated with diverse real-world decision problems including, but not limited to, sports, medicine, forensics, the UK housing market, and the UK financial market.
All applicants should hold, or close to completing, an MSc degree (or BSc with relevant experience) in an area related to computer science, statistics, or mathematics. Applicants with advanced knowledge in Bayesian methods, or with experience in publishing conference/journal publications, are particularly encouraged to apply. Strong motivation to aim for excellence is essential, as are excellent communication skills.
Applicants seeking further information or feedback on their suitability are encouraged to contact Dr. Anthony Constantinou (www.constantinou.info) at [email protected]
with subject “Bayesian-AI PhD”. Please attach your CV, a transcript of records, your BSc/MSc dissertation/s, and any conference/journal publications.
All nationalities are eligible to apply for this studentship. We offer a 3-years fully funded PhD studentship, with a tax-free bursary currently £16.8K/year for 2018/19, and a fee waiver (including non-EU students) supported by the School of Electronic Engineering and Computer Science (EECS) of the Queen Mary University of London, UK (www.eecs.qmul.ac.uk). The successful applicant will join EECS that has more than 300 PhDs, and will become a member of the Bayesian Artificial Intelligence Research Lab, and a member of the Risk and Information Management (RIM) research group.
please follow the on-line process at http://www.eecs.qmul.ac.uk/phd/how-to-apply/
. Please note that instead of a ‘Research Proposal’, we request a ‘Statement of Research Interests’. Your statement (no more than 500 words) should answer two questions: a) Why are you interested in the topic described above?, and b) What relevant experience do you have? In addition, we would also like you to send a sample of your written work. This might be your BSc or MSc dissertation, or a published conference or journal paper.
The closing date for the applications is March 15, 2019.
Interviews will start before the deadline and continue shortly after the deadline.
Starting date: As soon as possible or by October 2019.