Autonomous history-aware decision-making with explanation and reasoning capabilities

There is growing interest towards the application of ML-based learning for self-adaptation. However, the self-* property “self-explanation” of self-adaptive and autonomous systems (SAS) has been neglected. This is paradoxical, as self-explanation is inevitably needed when using ML techniques. 

In the SEA Research group, we are constructing our own infrastructure for SAS that use ML and Bayesian Learning (BL) to be able to (i) look at its own history to explain why the system has reached its current state and (ii) improve its decision making. The infrastructure and capabilities need to be built in such a way that the system's history can be stored and queried to be used in the context of the decision-making algorithms. The infrastructure is open source and managed by SEA, what allows external collaborations. The explanation capabilities are framed in four levels, (1) forensic history-aware explanation, (2) live history-aware explanation, (3) human-in-the-loop and (4) autonomous history-aware explanation. So far, 3 PhD students (one graduated) have constructed implementations for levels 1,2 and 3, with several international conferences and journal publications.

The aim of the PhD project is to develop level 4, i.e history-aware explanation capabilities to support autonomous behaviour of SAS that use ML and BL. The improved version of the existing decision-making process will take control over its history as another dimension to adapt and offer explanation of the behaviour. The specific objectives are:

• include other learning techniques not explored so far. These techniques build up on data structures, such as Q-tables in the Q-Learning model-free reinforcement learning algorithms, which have been used at level 2 and 3.
• explore links from deeper nuances of decision-making data with the time series modelling.

The student will work in the dynamic environment of SEA, interacting with other students, under the umbrella of the new Twenty20Insight EPSRC project.

Person Specification

The successful applicant should have been awarded, or expect to achieve, a Masters degree in a relevant subject with a 60% or higher weighted average, and/or a First or Upper Second Class Honours degree (or an equivalent qualification from an overseas institution) in Computer Science or equivalent. Preferred skill requirements include:

* Required: strong programming and software development skills

* Desirable: experience with engagement in open-source software projects

* Desirable: experience with machine learning approaches (especially Bayesian and Reinforcement Learning approaches) 

Submitting an application

As part of the application, you will need to supply: 

• A copy of your current CV 
• Copies of your academic qualifications for your Bachelor degree, and Masters degree; this should include both certificates and transcripts, and must be translated in to English 
• A research proposal statement* 
• Two academic references 
• Proof of your English Language proficiency 

Details of how to submit your application, and the necessary supporting documents can be found here. 

*The application must be accompanied by a “research proposal” statement. An original proposal is not required as the initial scope of the project has been defined, candidates should take this opportunity to detail how their knowledge and experience will benefit the project and should also be accompanied by a brief review of relevant research literature.

Please include the supervisor name, project title, and project reference in your Personal Statement.