ENDS: Explainability of Non-Deterministic Solvers
Robert Gordon University, University of Stirling,
British Telecom, ARR Craib & The Data Lab
Closing Date: 12 noon Friday 10th April 2020
Applications are sought for Research Studentships (PhD) in Computational Intelligence at Robert Gordon University (RGU) and at University of Stirling (UoS).
This proposal is a cutting-edge investigation into explaining the decisions of commonly-used solvers for optimization problems. Two PhD students, one at RGU and one at UoS, will be working on theoretical problems as well as with real-world case studies furnished by our industrial partners, BT and ARR Craib Ltd. This will exploit existing expertise and collaboration between the academic and industrial partners, with the potential for a step-change advance in how powerful optimization algorithms can be effectively harnessed to support human decision-makers in solving challenging industrial optimisation problems.
Explainable AI is a well-established concept, but research success in the area has mainly focused on methods that mimic human reasoning, making the path to solution readily understood by end-users. In non-deterministic solvers, the path to solution is driven by random processes that accumulate problem learning as they solve, as opposed to deduction from prior knowledge or experience. A technical description of these processes, while in some sense explanatory, is hard for non-experts to comprehend. The research challenge that ENDS will address is how to derive human-understandable knowledge about the problem from the non-deterministic solution process and translate that into an explanatory form for end-users.
Two distinct approaches, mining algorithm trajectories (RGU) and surrogate problem models (UoS), will investigate new ways to generate user-understandable problem knowledge from analysis of algorithm behaviour. Each ENDS PhD will also explore ways of using natural language generation and visualisation to translate the problem knowledge gained into explanations comprehensible to domain experts. End-users with no technical understanding of the solvers will thus be able to assess presented solutions in the light of a comprehensible explanation in the language of the application domain. The project will benefit throughout from exposure to two real world domains where non-deterministic solvers are already generating decision recommendations: workforce management at BT and real time truck scheduling at ARR Craib.
One student will be based at RGU under the supervision of Prof. McCall, working as a member of the Computational Intelligence research group. The other student will be based at UoS under the supervision of Dr. Brownlee, working in the Data Science research group. Both students will periodically visit the BT research facility at Adastral Park, Ipswich, working with the BT research team and applying their research to BT datasets. ARR Craib will provide operational data and end-user feedback to both students.
The studentships will be of 36 months duration, commencing in September 2020. The studentships are fully funded and include Home/EU tuition fees as well as a tax-free stipend of £15,000 per annum. Non-EU students may also apply but will be required to fund the difference between Home/EU fees and International fees.
Applicants should have, or expect to obtain before September 2020, a first class Honours degree or a Masters at Distinction level in Computing Science or a strongly related discipline. Strong programming skills are highly desirable. Some knowledge of non-deterministic optimization algorithms, in particular population-based techniques such as genetic algorithms, is highly desirable, but not essential. Applicants should have excellent personal and communication skills, strong professionalism and integrity and be confident working on their own initiative.
Applications can be emailed to email@example.com any time before 12 noon on Friday 10th April 2020. The applications should consist of a covering letter or personal statement of interest and a CV. Further information such as passport details or transcripts may be requested during the short-listing stage. Interviews will take place at RGU in Aberdeen in the week commencing 27th April 2020.
All enquiries should be addressed jointly to:
Prof. John McCall, firstname.lastname@example.org and
Dr. Alexander Brownlee, email@example.com